The A.I Megathread (LLM , GPT , Development)

bnew · Feb 5, 2025

Stacy Mitchell (@stacyfmitchell.bsky.social)

"DeepSeek is the canary in the coal mine. It’s warning us that when there isn’t enough competition, our tech industry grows vulnerable to its Chinese rivals, threatening U.S. geopolitical power in the 21st century." Lina Khan on the dangers of not breaking the market power of the Tech giants...

bsky.app

1/1
Stacy Mitchell

"DeepSeek is the canary in the coal mine. It’s warning us that when there isn’t enough competition, our tech industry grows vulnerable to its Chinese rivals, threatening U.S. geopolitical power in the 21st century."

Lina Khan on the dangers of not breaking the market power of the Tech giants:

bafkreie5r73ywag5nazq2pqw3n5msjrsxtlanqmcmkfvqfoo7ccelk5w2i@jpeg

To post tweets in this format, more info here: https://www.thecoli.com/threads/tips-and-tricks-for-posting-the-coli-megathread.984734/post-52211196

Alexander Reid Ross (@areidross.bsky.social)

"For years now, these companies have been arguing that the government must protect them from competition to ensure that America stays ahead." But I thought Musk wanted total deregulation! Anyway, here's Lina Khan with a sharp piece...

bsky.app

1/1
Alexander Reid Ross

"For years now, these companies have been arguing that the government must protect them from competition to ensure that America stays ahead."

But I thought Musk wanted total deregulation! Anyway, here's Lina Khan with a sharp piece...

Opinion | DeepSeek Serves as a Warning About Big Tech

bafkreiffic7mluoqfjdtrn3dmzduubzxbbu2uft4eco5xh27wu5qblxdke@jpeg

To post tweets in this format, more info here: https://www.thecoli.com/threads/tips-and-tricks-for-posting-the-coli-megathread.984734/post-52211196

Dare Obasanjo (@carnage4life) on Threads

Lina Khan is a Kendrick-level hater when it comes to big tech. 🤣

www.threads.net

Dare Obasanjo (@carnage4life) on Threads

Lina Khan is a Kendrick-level hater when it comes to big tech. 🤣

www.threads.net

1/14
@carnage4life
Lina Khan is a Kendrick-level hater when it comes to big tech.

Opinion | DeepSeek Serves as a Warning About Big Tech

2/14
@jimhong
Lina Khan is proof not all superheroes wear capes

3/14
@mcvalor.art
Love it man!!!

4/14
@mattismy1stname
If she was a Kendrick level. Hater then she would have actually made some hits and not duds. Maybe she's the Drake in this

5/14
@chasejacksonvibes
She was right. The facade is off now and we can see their true colors.

6/14
@theinstantwin
Everything she wrote here is correct. No notes.

7/14
@anindyabd
Strange arguments in this article -- did she forget it was Biden administration policy to enact the chip ban and TikTok ban in the name of national security? And does she think big tech and VCs like Andreessen (who says he represents "small tech") represent the same interests? The only major takeaway from this article is that she hates big tech but doesn't actually care about what tech companies do or how they compete with each other tooth and nail

8/14
@raytray4
So are most of the commenters on social media.

9/14
@documentingmeta
imagine not only driving your biggest, richest and most influential supporters to the hands of Trump but doubling down on your position after losing the election.
and all this when you had 4 years to do something, did jack shyt, then using your influence to write tech hit pieces on NYT. Lina Khan is incredible.

10/14
@tb_99999
Should we normalize just taking zooms from your desk, like a 1980s trading floor? It would be loud and chaotic, but the efficiency gains from not running between various rooms might be worth it.

11/14
@crumbler
I realize this is now the standard UI, but I truly don't understand how an average person is supposed to decide which model to use for which task

12/14
@documentingmeta
Meta has repurchased $129BN of stock in the past five years at an average price of $269. It is now trading at $704 $META

13/14
@darkzuckerberg
I’m back from threads jail, after a rate limiting bug on Instagram comments sent me there.
Don’t tag big creators who don’t follow you in other posts more than 3 times in a day or two, or you’ll get labeled as spam.
There’s no way to ask @zuck for a pardon via rage shake though lol

14/14
@joannastern
Been 24 hours and still no Apple Invites for me. It’s high school all over again.

To post threads in this format, more info here: https://www.thecoli.com/threads/tips-and-tricks-for-posting-the-coli-megathread.984734/post-52211196

Heimdall · Feb 5, 2025

Google owner drops promise not to use AI for weapons

Alphabet guidelines no longer refer to not pursuing technologies that could ‘cause or are likely to cause overall harm’

www.theguardian.com

The Google owner, Alphabet, has dropped its promise not to use artificial intelligence for purposes such as developing weapons and surveillance tools.

The US technology company said on Tuesday, just before it reported lower-than-forecast earnings, that it had updated its ethical guidelines around AI, and they no longer referred to not pursuing technologies that could “cause or are likely to cause overall harm”.

Google’s AI head, Demis Hassabis, said the guidelines were being overhauled in a changing world and that AI should protect “national security”.

In a blogpost defending the move, Hassabis and the company’s senior vice-president for technology and society, James Manyika, wrote that as global competition for AI leadership increased, the company believed “democracies should lead in AI development” that was guided by “freedom, equality, and respect for human rights”.

bnew · Feb 5, 2025

Bluesky

bsky.app

1/23
New York Times Opinion

DeepSeek is “warning us that when there isn’t enough competition, our tech industry grows vulnerable to its Chinese rivals, threatening U.S. geopolitical power in the 21st century,” writes Lina Khan, chair of the Federal Trade Commission in the Biden administration.

bafkreiedxqyprvzbqoi6g3cnfllfylx76gc2hrnsuljce6txvavjezzozy@jpeg

2/23
‪Sam Belman

‬ ‪@sambelman.bsky.social‬

DeepSeek serves as a warning?

But ChatGPT doesn't?

3/23
‪Comrade Eric‬ ‪@comradeeric.bsky.social‬

No it’s because Americas tech industry is a scam. Just raising massive amounts of money on speculation. Then China just proved it wasn’t necessary

4/23
‪The Compiler‬ ‪@sixnein.bsky.social‬

Seems like the least of our problems now days doesn't NYT?

5/23
‪Jan Dehn‬ ‪@jandehn.bsky.social‬

BLOG: Here is a cheap and smart way to introduce competition by design. Big Tech will hate it, but consumers will love it. And it will save the government tons of money in law suits.

How To Tackle Big E-commerce Monopolies Through Competition

bafkreibymbfo4fcdgmheufwsilstk6wftowj5luaxbehpsajfc7is3nvay@jpeg

6/23
‪Debra Walker‬ ‪@dwalker.bsky.social‬

Throw Musky in prison.
Problem solved.

7/23
‪Vincent Pyrmont‬ ‪@vincentpyrmont.bsky.social‬

Doesn't it warn, that big tech is killing the climate?

8/23
‪Peat blogg‬ ‪@manalive24.bsky.social‬

So DeepSeek is not regarded as competition?

9/23
‪Morning Coffee‬ ‪@jpx99.bsky.social‬

Hi ho silver lining..

10/23
‪msjaneoly.bsky.social‬ ‪@msjaneoly.bsky.social‬

I am done with you!!! You are tRump lovers.

11/23
‪Michael Samonek‬ ‪@samonek.bsky.social‬

"U.S. Geopolitical power" LOLOLOLOLOL

12/23
‪Robin Lud‬ ‪@fleshoversand.bsky.social‬

Because domination of global tech by the US is great and healthy for everybody.

13/23
‪roygbiv‬ ‪@bocuma.online‬

Once it gets to the point where energy infrastructure is the most immediate bottleneck, China will be in the driver’s seat. They are going to be leading the way in efficient, renewable energy while we are still jostling to drill in Greenland.

14/23
‪BAK88‬ ‪@greatlakes88.bsky.social‬

China made me lose faith in the market. I feel naïve now, because as China is not doing as well, the free market countries, particularly those who take the edges off of capitalism are doing well. Thinking of Scandinavia.

15/23
‪BAK88‬ ‪@greatlakes88.bsky.social‬

It’s funny because tech successfully convinced us that they needed to be big to compete with China.

16/23
‪Andy Schauer‬ ‪@schauer.bsky.social‬

Absolutely brain dead takeaway lol

17/23
‪Themistocles Wall‬ ‪@themistocleswall.bsky.social‬

1 of the few pieces in New York Times Opinion (@nytopinion.nytimes.com) I’d actually read if I still had a subscription

but the cowardice is too great The New York Times (@nytimes.com)

Don’t pay for cowardice

Themistocles Wall (@themistocleswall.bsky.social)

bafkreifukpfbgvfefg4z5mz5i3u4wyvazoaxgoouewtaaxdmmvruxwmug4@jpeg

18/23
‪ModernOddity‬ ‪@modernoddity.bsky.social‬

The same FTC who has been letting giant tech companies farm our data, whether it's Meta or your Cell service provider? Yeah, they can go fukk themselves.

19/23
‪Rune‬ ‪@runefar.bsky.social‬

More competition would be great but often people actually support monopoly unintentionally. Look how often people describe how they wish there wasnt multiple platforms for movies and just one. Despite this being understandable due to rising media costs, that ultimately is a monopoly

20/23
‪Rune‬ ‪@runefar.bsky.social‬

Additionally more people need to be made aware that it isnt just big tech designing things. There are a lot of other player who are much smaller and yet are still attempting to competite. More awareness of those individuals would be benefitial.

21/23
‪Rune‬ ‪@runefar.bsky.social‬

Finally the truth is that we really are on a infrastructure level behind on tech compared to other countries. This should be updated even if associated now with malicious actors yet it should also be combines with plans to counter inequity

22/23
‪yaboi kusama‬ ‪@lexaproletarian.bsky.social‬

State monopoly capitalism 101

23/23
‪bullxead‬ ‪@bullxead.bsky.social‬

About what?
Too late.
It was his job?

To post tweets in this format, more info here: https://www.thecoli.com/threads/tips-and-tricks-for-posting-the-coli-megathread.984734/post-52211196

bnew · Feb 5, 2025

Hugging Face researchers aim to build an 'open' version of OpenAI's deep research tool | TechCrunch

A group of developers at Hugging Face say that they've built an 'open' version of OpenAI's deep research tool.

techcrunch.com

Hugging Face researchers aim to build an ‘open’ version of OpenAI’s deep research tool

Kyle Wiggers

12:32 PM PST · February 4, 2025

A group of developers at AI dev platform Hugging Face, including Thomas Wolf, the company’s co-founder and chief scientist, say they’ve built an “open” version of OpenAI’s deep research tool.

Deep research, which OpenAI unveiled during an event Sunday, crawls the web to compile research reports on any subject. While impressive, deep research is currently only available in limited preview to users subscribed to OpenAI’s $200-a-month ChatGPT Pro plan.

The Hugging Face team’s project, which they’re calling Open Deep Research, consists of an AI model — OpenAI’s o1 — and an open source “agentic framework” that helps the model plan its analysis and guides it to use tools like search engines. O1 is a proprietary model (i.e. gated behind a paid API), but the team says it delivered better performance than “open” models such as DeepSeek’s R1.

In less than 24 hours, the researchers were able to harness o1 to use a simple, text-based browser and a “text inspector” toolkit to read files across the web. Open Deep Research can navigate the web autonomously, the team says, scrolling through pages, manipulating files, and even running calculations with data.

On GAIA, a benchmark for general AI assistants, Open Deep Research achieves a score of 54%. That’s compared with OpenAI deep research’s score of 67.36%.

I tried Open Deep Research in the public demo the team set up — but couldn’t get it to work. The page was under heavy load at publication time; after 10 minutes, it spit out an error message.

But the researchers say that they’re committed to improving the experience, and have made the source code available on GitHub for inspection and feedback.

Worth noting is that there are a number of OpenAI deep research “reproductions” on the web, some of which rely on open models and tooling. The crucial component they — and Open Deep Research — lack is o3, the model underpinning deep research.

Few, if any, models beat o3 on benchmarks related to answering complex questions and information gathering. Short of an open model to rival o3, deep research alternatives may not quite measure up to the real thing.

bnew · Feb 5, 2025

Google launches new AI models and brings 'thinking' to Gemini | TechCrunch

Google has launched several new AI models, including Gemini 2.0 Pro Experimental, and brought a 'reasoning' model to its Gemini app.

techcrunch.com

Google launches new AI models and brings ‘thinking’ to Gemini

Maxwell Zeff

8:00 AM PST · February 5, 2025

Google launched its much-anticipated new flagship AI model, Gemini 2.0 Pro Experimental, on Wednesday. The announcement was part of a series of other AI model releases. The company is also making its “reasoning’ model, Gemini 2.0 Flash Thinking, available in the Gemini app.

Notably, Google is releasing these AI models as the tech world remains fixated on cheaper AI reasoning models offered by the Chinese AI startup DeepSeek. DeepSeek’s models match or surpass the performance of leading AI models offered by American tech companies. At the same time, businesses can access DeepSeek’s models through the company’s API for a relative steal.

Google and DeepSeek both released AI reasoning models in December, but DeepSeek’s R1 got a lot more attention. Now, Google may be trying to put its Gemini 2.0 Flash Thinking model in front of more eyes through its popular Gemini app.

As for Gemini 2.0 Pro, the successor to the Gemini 1.5 Pro model Google launched last February, Google says that it is now the leading model in its Gemini AI model family.

Google accidentally announced the Gemini 2.0 Pro model’s release in the Gemini app’s changelog roughly a week ago. But this time, it’s for real. The company is releasing an experimental version of the model on Wednesday in its AI development platforms, Vertex AI and Google AI Studio. Gemini 2.0 Pro will also be available to subscribers to Gemini Advanced in the Gemini app.

Specifically, the new Gemini Pro model excels at coding and handling complex prompts, per Google, and it comes with “better understanding and reasoning of world knowledge” than any of the company’s previous models. Gemini 2.0 Pro can call tools like Google Search, and execute code on behalf of users.

Gemini 2.0 Pro’s context window is 2 million tokens, meaning it can process about 1.5 million words in one go. Expressed another way, Google’s newest AI model could ingest all seven books in the Harry Potter series in a single prompt and still have about 400,000 words left over.

Google is also making its Gemini 2.0 Flash model generally available on Wednesday. This model was announced in December and is now available to all users of the Gemini app.

Lastly, possibly to rival the excitement surrounding DeepSeek’s models, Google is introducing a new, more cost-efficient AI model, Gemini 2.0 Flash-Lite. The company says this model outperforms its Gemini 1.5 Flash model, but runs at the same price and speed.

\

bnew · Feb 5, 2025

Researchers created an open rival to OpenAI's o1 'reasoning' model for under $50 | TechCrunch

AI researchers at Stanford and the University of Washington were able to train an AI "reasoning" model for under $50 in cloud compute credits, according

techcrunch.com

Researchers created an open rival to OpenAI’s o1 ‘reasoning’ model for under $50

Maxwell Zeff

3:38 PM PST · February 5, 2025

AI researchers at Stanford and the University of Washington were able to train an AI “reasoning” model for under $50 in cloud compute credits, according to a new research paper released last Friday.

The model known as s1 performs similarly to cutting-edge reasoning models, such as OpenAI’s o1 and DeepSeek’s R1, on tests measuring math and coding abilities. The s1 model is available on GitHub, along with the data and code used to train it.

The team behind s1 said they started with an off-the-shelf base model, then fine-tuned it through distillation, a process to extract the “reasoning” capabilities from another AI model by training on its answers.

The researchers said s1 is distilled from one of Google’s reasoning models, Gemini 2.0 Flash Thinking Experimental. Distillation is the same approach Berkeley researchers used to create an AI reasoning model for around $450 last month.

To some, the idea that a few researchers without millions of dollars behind them can still innovate in the AI space is exciting. But s1 raises real questions about the commoditization of AI models.

Where’s the moat if someone can closely replicate a multi-million dollar model with relative pocket change?

Unsurprisingly, big AI labs aren’t happy. OpenAI has accused DeepSeek of improperly harvesting data from its API for the purposes of model distillation.

The researchers behind s1 were looking to find the simplest approach to achieve strong reasoning performance and “test-time scaling,” or allowing an AI model to think more before it answers a question. These were a few of the breakthroughs in OpenAI’s o1, which DeepSeek and other AI labs have tried to replicate through various techniques.

The s1 paper suggests that reasoning models can be distilled with a relatively small dataset using a process called supervised fine-tuning (SFT), in which an AI model is explicitly instructed to mimic certain behaviors in a dataset.

SFT tends to be cheaper than the large-scale reinforcement learning method that DeepSeek employed to train its competitor to OpenAI’s o1 model, R1.

Google offers free access to Gemini 2.0 Flash Thinking Experimental, albeit with daily rate limits, via its Google AI Studio platform.

Google’s terms forbid reverse-engineering its models to develop services that compete with the company’s own AI offerings, however. We’ve reached out to Google for comment.

S1 is based on a small, off-the-shelf AI model from Alibaba-owned Chinese AI lab Qwen, which is available to download for free. To train s1, the researchers created a dataset of just 1,000 carefully curated questions, paired with answers to those questions as well as the “thinking” process behind each answer from Google’s Gemini 2.0 Flash Thinking Experimental.

After training s1, which took less than 30 minutes using 16 Nvidia H100 GPUs, s1 achieved strong performance on certain AI benchmarks, according to the researchers. Niklas Muennighoff, a Stanford researcher who worked on the project, told TechCrunch he could rent the necessary compute today for about $20.

The researchers used a nifty trick to get s1 to double-check its work and extend its “thinking” time: they told it to wait. Adding the word “wait” during s1’s reasoning helped the model arrive at slightly more accurate answers, per the paper.

In 2025, Meta, Google, and Microsoft plan to invest hundreds of billions of dollars in AI infrastructure, which will partially go toward training next-generation AI models.

That level of investment may still be necessary to push the envelope of AI innovation. Distillation has shown to be a good method for cheaply recreating an AI model’s capabilities, but it doesn’t create new AI models vastly better than what’s available today.

bnew · Feb 5, 2025

Naver-backed Cinamon wants to make 3D video animation easier using AI | TechCrunch

It's never been easier to create and publish art than it is now, and if you believe the companies building tech around AI, the production process is going

techcrunch.com

Naver-backed Cinamon wants to make 3D video animation easier using AI

Kate Park

3:00 PM PST · February 3, 2025

It’s never been easier to create and publish art than it is now, and if you believe the companies building tech around AI, the production process is going to get even more efficient. That’s especially the case with video production, with companies of all sizes using large language models to build tools that let you whip up decent-quality videos and animation with a few prompts and actions.

Popular tools in this space include Google’s Veo 2, OpenAI’s Sora, Runway, Luma AI, and Shanghai-based Hailuo. Now, a South Korean startup called Cinamon is ramping up efforts to claim a part of this burgeoning market — it recently raised an $8.5 million to continue building its animated video generation platform “CINEV,” slated to be launched in beta in the first half of 2025. Altos Ventures, an existing backer, as well as Saehan Venture Capital invested in this round.

Cinamon’s pitch is that its platform provides a video generator that can let you build 3D environments, direct scenes and actions, place characters, edit camera angles, and more — all with text prompts and sliders.

According to its CEO Doosun Hong, the company’s approach fundamentally differs from existing AI video generators, which create videos by generating pixels using text, images, and videos as reference materials. In contrast, CINEV combines a 3D asset library, AI motion generation, and a filmmaking-focused large language model to first construct 3D scenes complete with characters and elements, and then lets you edit them using its suite of video production and editing tools.

Some of the video generation and editing tools in CINEV. Image Credits: CinamonImage Credits:Cinamon

“Our approach enables easier direction and editing without consistency/physics issues, making it particularly suitable for longer-form content like films and dramas,” Hong said. “We envision CINEV to be complementary to existing AI video tools, potentially enabling new workflows where CINEV’s output could serve as high-quality reference material for other AI video platforms.”

Cinamon started life in 2019 as Cinamon Games, a subsidiary of content production firm Vonvon. Cinamon initially set up a JV with NAVER WEBTOON, a Korean digital storytelling platform, to create Maybe, an interactive storytelling app. Facing growing data privacy concerns, Vonvon merged with Cinamon Games later in 2019 to focus solely on storytelling in the social content space.

While its competitors like Crazy Maple Studio started offering animated interactive story apps, fiction apps, storytelling apps, and short-form videos, Cinamon instead chose to focus on creating 3D animation tools that could speed up and scale animation production for content creators and studios. Even though it cost more investment than tools for 2D content, they saw greater scalability potential.

In 2022, the startup began building its 3D animation platform, and later integrated AI features to enhance production efficiency. In September that year, South Korean gaming company Krafton, Naver Z (a unit of Korean internet giant Naver), and SNOW, (a camera app run by Naver), invested $10 million in Cinamon’s Series A.

Going forward, Cinamon plans to use its investors’ IPs and 3D assets to bolster its offering. Krafton has substantial intellectual property and 3D assets used in its Battleground games, while Naver Z operates the Zepeto metaverse platform. Cinamon says CINEV can help extend these IPs beyond gaming by enabling content creators to use these IPs, and potentially drive user acquisition for Krafton and Naver Z. Cinamon also entered Nvidia’s startup accelerator, Inception, last August.

“Our potential users include comic, manga, webtoon artists, web novel writers, game developers, video creators, and traditional animators looking for easier workflows,” Hong said. “In 2025, we plan to focus on customers [ranging] from individual creators to content IP companies looking for easier, faster, cheaper ways to create Anime, VTuber, and cinematic video game content.”

Cinamon plans to use the new capital for hiring more AI engineers and for R&D. The startup has a team of 60 staff with expertise in 3D graphics, AI, gaming, and content production. The latest funding brings its total capital raised so far to $18.5 million (25 billion KRW).

bnew · Feb 5, 2025

Meta says it may stop development of AI systems it deems too risky | TechCrunch

Meta has released a policy document outlining scenarios in which the company may not release certain categories of 'risky' AI systems.

techcrunch.com

Meta says it may stop development of AI systems it deems too risky

Kyle Wiggers

1:52 PM PST · February 3, 2025

Meta CEO Mark Zuckerberg has pledged to make artificial general intelligence (AGI) — which is roughly defined as AI that can accomplish any task a human can — openly available one day. But in a new policy document, Meta suggests that there are certain scenarios in which it may not release a highly capable AI system it developed internally.

The document, which Meta is calling its Frontier AI Framework, identifies two types of AI systems the company considers too risky to release: “high risk” and “critical risk” systems.

As Meta defines them, both “high-risk” and “critical-risk” systems are capable of aiding in cybersecurity, chemical, and biological attacks, the difference being that “critical-risk” systems could result in a “catastrophic outcome [that] cannot be mitigated in [a] proposed deployment context.” High-risk systems, by contrast, might make an attack easier to carry out but not as reliably or dependably as a critical risk system.

Which sort of attacks are we talking about here? Meta gives a few examples, like the “automated end-to-end compromise of a best-practice-protected corporate-scale environment” and the “proliferation of high-impact biological weapons.” The list of possible catastrophes in Meta’s document is far from exhaustive, the company acknowledges, but includes those that Meta believes to be “the most urgent” and plausible to arise as a direct result of releasing a powerful AI system.

Somewhat surprising is that, according to the document, Meta classifies system risk not based on any one empirical test but informed by the input of internal and external researchers who are subject to review by “senior-level decision-makers.” Why? Meta says that it doesn’t believe the science of evaluation is “sufficiently robust as to provide definitive quantitative metrics” for deciding a system’s riskiness.

If Meta determines a system is high-risk, the company says it will limit access to the system internally and won’t release it until it implements mitigations to “reduce risk to moderate levels.” If, on the other hand, a system is deemed critical-risk, Meta says it will implement unspecified security protections to prevent the system from being exfiltrated and stop development until the system can be made less dangerous.

Meta’s Frontier AI Framework, which the company says will evolve with the changing AI landscape, and which Meta earlier committed to publishing ahead of the France AI Action Summit this month, appears to be a response to criticism of the company’s “open” approach to system development. Meta has embraced a strategy of making its AI technology openly available — albeit not open source by the commonly understood definition — in contrast to companies like OpenAI that opt to gate their systems behind an API.

For Meta, the open release approach has proven to be a blessing and a curse. The company’s family of AI models, called Llama, has racked up hundreds of millions of downloads. But Llama has also reportedly been used by at least one U.S. adversary to develop a defense chatbot.

In publishing its Frontier AI Framework, Meta may also be aiming to contrast its open AI strategy with Chinese AI firm DeepSeek’s. DeepSeek also makes its systems openly available. But the company’s AI has few safeguards and can be easily steered to generate toxic and harmful outputs.

“[W]e believe that by considering both benefits and risks in making decisions about how to develop and deploy advanced AI,” Meta writes in the document, “it is possible to deliver that technology to society in a way that preserves the benefits of that technology to society while also maintaining an appropriate level of risk.”

bnew · Feb 6, 2025

Image Credits:ByteDance

Deepfake videos are getting shockingly good

Kyle Wiggers

8:18 AM PST · February 4, 2025

Researchers from TikTok owner ByteDance have demoed a new AI system, OmniHuman-1, that can generate perhaps the most realistic deepfake videos to date.

Deepfaking AI is a commodity. There’s no shortage of apps that can insert someone into a photo, or make a person appear to say something they didn’t actually say. But most deepfakes — and video deepfakes in particular — fail to clear the uncanny valley. There’s usually some tell or obvious sign that AI was involved somewhere.

Not so with OmniHuman-1 — at least from the cherry-picked samples the ByteDance team released.

Here’s a fictional Taylor Swift performance. Here’s a TED Talk that never took place. And here’s a deepfaked Einstein lecture:

According to the ByteDance researchers, OmniHuman-1 only needs a single reference image and audio, like speech or vocals, to generate a clip of an arbitrary length. The output video’s aspect ratio is adjustable, as is the subject’s “body proportion” — i.e. how much of their body is shown in the fake footage.

Trained on 19,000 hours of video content from undisclosed sources, OmniHuman-1 can also edit existing videos — even modifying the movements of a person’s limbs. It’s truly astonishing how convincing the result can be.

Granted, OmniHuman-1 isn’t perfect. The ByteDance team says that “low-quality” reference images won’t yield the best videos, and the system seems to struggle with certain poses. Note the weird gestures with the wine glass in this video:

Still, OmniHuman-1 is easily heads and shoulders above previous deepfake techniques, and it may well be a sign of things to come. While ByteDance hasn’t released the system, the AI community tends not to take long to reverse-engineer models like these.

The implications are worrisome.

Last year, political deepfakes spread like wildfire around the globe. On election day in Taiwan, a Chinese Communist Party-affiliated group posted AI-generated, misleading audio of a politician throwing his support behind a pro-China candidate. In Moldova, deepfake videos depicted the country’s president, Maia Sandu, resigning. And in South Africa, a deepfake of rapper Eminem supporting a South African opposition party circulated ahead of the country’s election.

Deepfakes are also increasingly being used to carry out financial crimes. Consumers are being duped by deepfakes of celebrities offering fraudulent investment opportunities, while corporations are being swindled out of millions by deepfake impersonators. According to Deloitte, AI-generated content contributed to more than $12 billion in fraud losses in 2023, and could reach $40 billion in the U.S. by 2027.

Last February, hundreds in the AI community signed an open letter calling for strict deepfake regulation. In the absence of a law criminalizing deepfakes at the federal level in the U.S., more than 10 states have enacted statutes against AI-aided impersonation. California’s law — currently stalled — would be the first to empower judges to order the posters of deepfakes to take them down or potentially face monetary penalties.

Unfortunately, deepfakes are hard to detect. While some social networks and search engines have taken steps to limit their spread, the volume of deepfake content online continues to grow at an alarmingly fast rate.

In a May 2024 survey from ID verification firm Jumio, 60% of people said they encountered a deepfake in the past year. Seventy-two percent of respondents to the poll said they were worried about being fooled by deepfakes on a daily basis, while a majority supported legislation to address the proliferation of AI-generated fakes.

bnew · Feb 8, 2025

Marc Love (@marcslove) on Threads

LLM companies like Anthropic & OpenAI are really caught in a catch 22. Their models are commoditized so quickly that they likely can’t even recover the cost of training them. The answer would seem to be to build defensible product around those models. The problem is their products are...

www.threads.net

Marc Love (@marcslove) on Threads

LLM companies like Anthropic & OpenAI are really caught in a catch 22. Their models are commoditized so quickly that they likely can’t even recover the cost of training them. The answer would seem to be to build defensible product around those models. The problem is their products are...

www.threads.net

Marc Love (@marcslove) on Threads

They're caught in a rapid cycle of innovation though and they aren't increasing their moat, they're losing it. If innovation slows down, it gives competitors a chance to catch up. If innovation speeds up, they're spending even more money for a rapidly depreciating asset. They may go into...

www.threads.net

Marc Love (@marcslove) on Threads

Hah! Well, depends on who you’re talking about. 😉 These companies have people all along the spectrum from researcher who cares almost exclusively about their research and craft to avaricious capitalists.

www.threads.net

Marc Love (@marcslove) on Threads

It’s a long term bet that AI products become as essential to running a business as buying a computer for each of your employees. I think that’s still several years away before they become that widespread and essential. I’m also skeptical that companies will be able to capture and protect that...

www.threads.net

Marc Love (@marcslove) on Threads

Oh I totally agree. Hardware's a far better investment in my opinion. No matter how commoditized LLMs or other GPU-accelerated models become, deep learning is here, incredibly valuable, and is going to "eat the word" like software did. Whether OpenAI becomes the largest company in the world or...

www.threads.net

Marc Love (@marcslove) on Threads

They face a challenge that's much more like software. They'll have to provide enough value above and beyond using free and open source solutions that enterprises would rather buy them "off the shelf" than build it themselves. My first instinct that it has to do with MLOps/LLMOps and managing the...

www.threads.net

Marc Love (@marcslove) on Threads

Congestion pricing is already a huge success. I wonder what % of people complaining about it would have happily forked over $9 to use an express lane that would cut their commute time in half.

www.threads.net

1/17
@marcslove
LLM companies like Anthropic & OpenAI are really caught in a catch 22.
Their models are commoditized so quickly that they likely can’t even recover the cost of training them.
The answer would seem to be to build defensible product around those models.
The problem is their products are intrinsically linked to their models. The commoditization of the model thus devalues the product.
That leaves an opening for model agnostic products to “cross their product moat.”

2/17
@mshonle
What do people in SV even mean when they say "moat"? I've only heard it described in the absence of one, so what company is an example of having a moat? Does Uber have a moat? Does Walmart? (If so, what are they, if not then does any company have a moat?)

3/17
@mark_r_vickers
That's true, and I think they are building decent products based on those models and releasing them quickly. And, they are also the engines behind countless wrappers that need to pay them. That's a low margin business but important because eventually AIs will become too powerful to open source. I think that eventually, like IBM in many companies before them, they may become consultants.

4/17
@marcslove
They're caught in a rapid cycle of innovation though and they aren't increasing their moat, they're losing it. If innovation slows down, it gives competitors a chance to catch up. If innovation speeds up, they're spending even more money for a rapidly depreciating asset.
They may go into consulting eventually. Less so like IBM and moreso like Hashicorp, Snowflake, Databricks, etc. as a platform with a large professional services business on top.

5/17
@mark_r_vickers
Yes, those are good analogies. On the other hand, this isn't just a business to them. It's a kind of calling, for better or for worse.

6/17
@marcslove
Hah! Well, depends on who you’re talking about.

These companies have people all along the spectrum from researcher who cares almost exclusively about their research and craft to avaricious capitalists.

7/17
@tsean_k
The estimated valuations on these companies seem out of line, even for the internet hyperbole era.
I can see some rational value to Nvidia.
But estimates of billions for these AI startups seems like hype to make the stocks go volcanic at IPO.

8/17
@marcslove
It’s a long term bet that AI products become as essential to running a business as buying a computer for each of your employees.
I think that’s still several years away before they become that widespread and essential.
I’m also skeptical that companies will be able to capture and protect that market long-term, especially if the product is really just compute + knowledge + math.
It’s a very different business than designing, building, and selling hardware.

9/17
@tsean_k
But what’s the revenue model? Cloud based subscription services? License buys with updates?
I lived through the era of ERPs and CRMs in business, so I can see that revenue models.
I still think the more predictable bet is the hardware guys, Nvidia is the long term play.

10/17
@marcslove
Oh I totally agree. Hardware's a far better investment in my opinion. No matter how commoditized LLMs or other GPU-accelerated models become, deep learning is here, incredibly valuable, and is going to "eat the word" like software did. Whether OpenAI becomes the largest company in the world or goes bankrupt, Nvidia's going to be trying to catch up to demand for years, if not decades. And their business is far more defensible.
As for the revenue model for LLMs long term…good question.

11/17
@marcslove
They face a challenge that's much more like software. They'll have to provide enough value above and beyond using free and open source solutions that enterprises would rather buy them "off the shelf" than build it themselves. My first instinct that it has to do with MLOps/LLMOps and managing the complexity of orchestration, monitoring, data pipelines, security, and compliance.

12/17
@joenandez
Dude we are so on the same wavelength some times.
Starting to convince myself that the future winning AI product(s) abstract away the underlying model and just figures out how to use the best model for the customer's use case, while rapidly innovating on the user experience itself.
Customers can trust they are always getting the optimal level of intelligence for the task, and don't have to tune it themselves.
Model providers relying on only their models have a strategic vulnerability.

[Quoted post]
joenandez
Joe Fernandez (@joenandez) on Threads

13/17
@joenandez
Right now, every new model that comes out developers are comparing/contrasting, and expecting their favorite AI Dev Tool to add access immediately.
And as we've seen from Deepseek's Appstore ranking, even consumers are not immune to model hopping,
This is a temporary phase in the AI Revolution ... so what's next?

14/17
@ociubotaru
I would gladly pay for a great voice assistant powered by sonnet 3.5

15/17
@jwynia
The Eleven Labs voice agents can be configured to use Sonnet as the LLM

16/17
@mark_r_vickers
It must be annoying to have one of the best and safest AIs and then watch all this hype and app downloads for a newby model where safety wasn't prioritized
techcrunch.com/2025…

Anthropic CEO says DeepSeek was 'the worst' on a critical bioweapons data safety test | TechCrunch
Anthropic CEO says DeepSeek was 'the worst' on a critical bioweapons data safety test | TechCrunch

17/17
@marcslove
Congestion pricing is already a huge success.
I wonder what % of people complaining about it would have happily forked over $9 to use an express lane that would cut their commute time in half.

[Quoted post]
alangbrake
alangbrake (@alangbrake) on Threads

https://www.fastcompany.com/9127243...is-like-after-one-month-of-congestion-pricing

alangbrake (@alangbrake) on Threads

We need to start trumpeting this, so even if it’s halted, we can revive it if we make it to the other side.

www.threads.net

18/19
@alangbrake
We need to start trumpeting this, so even if it’s halted, we can revive it if we make it to the other side.

https://www.fastcompany.com/9127243...is-like-after-one-month-of-congestion-pricing

19/19
@carnage4life
Censorship is relative. Many people were quick to point out DeepSeek won’t talk about “Tank man” and Tiananmen Square, now there are similar complaints that DeepSeek doesn’t censor content that American AI models do.
From hate speech to how to make weapons, DeepSeek will tell you things ChatGPT won’t.

https://www.wsj.com/tech/ai/china-deepseek-ai-dangerous-information-e8eb31a8

To post threads in this format, more info here: https://www.thecoli.com/threads/tips-and-tricks-for-posting-the-coli-megathread.984734/post-52211196

bnew · Feb 8, 2025

https://archive.is/mkwm5

bnew · Feb 8, 2025

1/11
@pika_labs
Wait…WHAT? Introducing Pikadditions, the easiest way to make your content stand out.

Add anyone or anything to any video, whether that’s a video you shoot yourself, or a favorite clip. Special surprise: get fifteen free Pikadditions generations when you sign up!

Go try it at pika dot art

https://video.twimg.com/ext_tw_video/1887541899302694912/pu/vid/avc1/1080x1350/ZkkVZnWgkCL18Jum.mp4

2/11
@Court_Reinland
I’ll have to take a Pika at this

3/11
@pika_labs
Go for it! Pikadditions is now live on Pika!

4/11
@Iamtomblake
Why do your magical things ship while I'm stuck in a meeting!?! It's a cruel world, lol. Looking forward to this!

5/11
@pika_labs
Hi Tom! It's never too late to join us

Thank you for supporting Pikadditions!

6/11
@SentientOne1
Awesome.
It's gonna unleash a barrage of fun videos on Instagram.

7/11
@pika_labs
Hope so!! Thank you for supporting Pikadditions

8/11
@BadIdeasForLife
Wow. So cool!

9/11
@pika_labs
Thank you

10/11
@HappyKittenAI
Amazing trailer!! Love the new feature!

11/11
@pika_labs
We have more on the way Irina!! Thank you so much for supporting Pikadditions! Your clip looks amazing in our launch video

To post tweets in this format, more info here: https://www.thecoli.com/threads/tips-and-tricks-for-posting-the-coli-megathread.984734/post-52211196

1/11
@CaptainHaHaa
Experimenting with some @pika_labs Pikadditions

https://video.twimg.com/ext_tw_video/1888264798154133504/pu/vid/avc1/1280x720/Ko5_7Dq_A9EDi_1d.mp4

2/11
@koldo2k

Awesome!

3/11
@CaptainHaHaa
Thanks Koldo it did not take very long at all to make

4/11
@GrimfelOfficial
Haha love this man <3

5/11
@CaptainHaHaa
Thanks mate, I was trying to find the right scene this one fit

️

6/11
@guicastellanos1
Soooo cool I didn't wee you when I watched, I have to watch it again!

7/11
@CaptainHaHaa
HAhaha I was cut due to budget issues

8/11
@botzero_net
LOL! This is really solid blending.

9/11
@CaptainHaHaa
remarkably easy

10/11
@AiDreamzX
One of my most favorite films ever!!

11/11
@CaptainHaHaa
Thank you mate me too!

To post tweets in this format, more info here: https://www.thecoli.com/threads/tips-and-tricks-for-posting-the-coli-megathread.984734/post-52211196

1/9
@Diesol
I had some fun on a Saturday morning with my boy and @pika_labs Pikadditions.

https://video.twimg.com/ext_tw_video/1888308301982765056/pu/vid/avc1/1080x1920/Xl5RtCSM7hYrW7f9.mp4

2/9
@ytjessie_
Brilliant! really scared me lol

3/9
@Diesol
Alien input image

4/9
@DeHavenAI
Even the dog barking sound effect is pretty cool

5/9
@Diesol
That was funny enough real and in camera ha, That was my greyhound

6/9
@Ror_Fly
epic

7/9
@azed_ai
Haha, that really scared me!
Amazing Dave

8/9
@guicastellanos1

9/9
@AmicaeAeternum

Brilliant

To post tweets in this format, more info here: https://www.thecoli.com/threads/tips-and-tricks-for-posting-the-coli-megathread.984734/post-52211196

1/7
@yuvalalaluf
Your morning commute will never be the same with Pikadditions!

Try it now with @pika_labs

https://video.twimg.com/ext_tw_video/1887578123706777600/pu/vid/avc1/1076x606/9wHUJb_aXQ6D2EaN.mp4

2/7
@omerbartal
Best way to travel to the office :smile:

3/7
@yuvalalaluf
We even have a new team member

[Quoted tweet]
Time to reveal the brains behind @pika_labs Pikadditions

https://video.twimg.com/ext_tw_video/1887577509572628481/pu/vid/avc1/478x850/Zb_UxOaMysYTZeN4.mp4

4/7
@chenlin_meng
Love the

created by Pikadditions.

5/7
@chenlin_meng
I need

team member

6/7
@Eddie395402252
super cool!

7/7
@minguk_kang
Is there a weight limit for that?

To post tweets in this format, more info here: https://www.thecoli.com/threads/tips-and-tricks-for-posting-the-coli-megathread.984734/post-52211196

bnew · Feb 9, 2025

https://archive.is/JYIqO

1/29
@WesRothMoney
OpenAI *coding* progress:
1st reasoning model = 1,000,000th best coder in the world
o1 (Sept 2024) was ranked = 9800th
o3 (Jan 2025) was ranked = 175th
(today) internal model = 50th

superhuman coder by eoy 2025?

https://video.twimg.com/ext_tw_video/1888330009334743040/pu/vid/avc1/1280x720/JLZCr6fUNW_SGNym.mp4

2/29
@WesRothMoney
I had to edit the tweet, I put 2023 as the date for some reason

/shrug

thanks to everyone who pointed that out :smile:

3/29
@WesRothMoney
here's the full video I did with all the highlights from that talk:

https://invidious.poast.org/4Wa6St-uosY

4/29
@mikeboysen
I wonder what the 50th best code or thinks. Has anybody interviewed him
Lol

5/29
@WesRothMoney
he's re-reading The Butlerian Jihad...

(jokes aside, I think the software engineers will benefit greatly from AI coding tools)

6/29
@circlerotator
competitive programming is more like competitive math than software engineering

something to keep in mind

7/29
@WesRothMoney
yeah, I don't think it 'replaces' great engineers.

I do think it will 'enable' great engineers.

8/29
@drjfhll
I still think anthropic is better; and Gemini catching up

9/29
@erdavtyan
Extremely tightly scoped problems with a lot of research and algo combinations published and trained on.

Superhuman coder should be able to work on complex, high-context systems that have multiple moving parts and legacy code. They should fix versioning / deployment issues.

10/29
@doeurlich50289
Hearing sama making such direct claims means they'll crush 2025, and by the end of the year, we'll enter a new world and have to accept a new reality.

11/29
@SulkaMike
A lot of interesting takes here, summarized around the question... Even if it's number one on the benchmark does that change much?

. And if does induce change, why doesn't 10 million people with a plus account and the 175th ranked prog have changed the world so far?

12/29
@OlivioSarikas
If it is that good, why does basically any coder I know tell me that AI is good at simple code, but as soon as it becomes more complex, writing the code yourself is faster than finding the AI errors in the code?

13/29
@rosdikuat
I'm quite certain this will happen by December. Even today I mostly don't code, I mostly prompt.

14/29
@JOSmithIII
Does anyone know where the o3-mini tiers rank?

15/29
@ImJayBallentine
“We have a superior coding model but we are just gonna let Sonnet keep the lead.” Got it.

16/29
@hagestev
what happened to o2??

17/29
@langdon
A single “best‐fit” exponential model through the three data points projects reaching Rank 1 around April-May 2025. The initial drop was extremely fast (Sept→Jan), while the more recent decline (Jan→Feb) was slower - so if you weigh later data more, you’d land closer to mid‐ or late Summer 2025.

18/29
@0xShawnWang
source of rank？

19/29
@DavidPrice21106
This is getting crazy, Wes.

20/29
@_oddfox_
Once these coding agents are out publicly shyt is really going to take off. Seems like 2026 is the year of the intelligence explosion

21/29
@3DTechPrep
What used to be the difficult part of my projects (code) is now the easy part.

So simple now and have learned more in the last year about coding than in past 20.

It’s like having a brilliant coder always there to ask ANY question, no matter how dumb or hard, no judgement.

22/29
@ArcherNightfall
How many times does sama have to say it. How many times.

23/29
@jfp618
Get testing score does is just one attribute of a good engineer

24/29
@unaliveolives
Try to build and maintain a real app with o3. It is, for sure, not the world’s 175th best coder.

25/29
@PaulMaddison121
Software engineering is solving problems not churning out syntax like LLMs do

For example the trillions of reasoning models needed for AIs growth will need software engineers to create/implement.

26/29
@VojtechKulhavy
Here's the plot :smile:

27/29
@wei_andrew
Why’s OpenAI still having many programmers?

28/29
@ChefBeijing
Most top researchers in OpenAI may not be senior software engineers on real world projects, which think programmer contest like a shyt. You need a poor guy from China or India to dig into 3000 files and each of them have 3000 or 5000 lines of code and variables to fix a bug

29/29
@keithofaptos
If OpenAI truly wants to be in the right side of History, it would be marvelous to receive this internal model (50th best global coder) ASAP and in voice2voice, completely open sourced. That's what us non coders are just itching to for. Imagine paying $20/m for this?! 🫠

@sama

To post tweets in this format, more info here: https://www.thecoli.com/threads/tips-and-tricks-for-posting-the-coli-megathread.984734/post-52211196

bnew · Feb 10, 2025

1/3
@_akhaliq
FlashVideo

Flowing Fidelity to Detail for Efficient High-Resolution Video Generation

https://video.twimg.com/ext_tw_video/1888826565078790144/pu/vid/avc1/1280x720/POLBuPlfFIXpr-h6.mp4

2/3
@_akhaliq
discuss: Paper page - FlashVideo:Flowing Fidelity to Detail for Efficient High-Resolution Video Generation

3/3
@UniversaAI
That sounds like an interesting project. The potential applications of high-resolution video generation are vast, from entertainment to education and beyond. How does FlashVideo's approach to efficient high-resolution video generation differ from other methods?

To post tweets in this format, more info here: https://www.thecoli.com/threads/tips-and-tricks-for-posting-the-coli-megathread.984734/post-52211196

bnew · Feb 10, 2025

Goku

Flow Based Video Generative Foundation Models

saiyan-world.github.io

GitHub - Saiyan-World/goku: Video Generation Foundation Models: https://saiyan-world.github.io/goku/

Video Generation Foundation Models: https://saiyan-world.github.io/goku/ - Saiyan-World/goku

github.com

1/11
@_akhaliq
ByteDance just dropped Goku

Flow Based Video Generative Foundation Models

https://video.twimg.com/ext_tw_video/1888811154744410112/pu/vid/avc1/1280x720/8vg8K8tBY83CKwe7.mp4

2/11
@_akhaliq
discuss: Paper page - Goku: Flow Based Video Generative Foundation Models

3/11
@_akhaliq
Dataset: saiyan-world/Goku-MovieGenBench · Datasets at Hugging Face

4/11
@noelhatem
Almost indistinguishable from real videos

5/11
@that_anokha_boy
dropped what?

6/11
@nickcammarata
it’s so over

7/11
@fifty7fifty7
they would have a nice database to learn from

8/11
@overac111ever
bytedance stop teasing me release it pls

9/11
@DKRacingFan

10/11
@ihteshamit
Goku?

it will be nice!

11/11
@MoonlitMonkey69
There's a pointless job that AI can replace: Influencers.

To post tweets in this format, more info here: https://www.thecoli.com/threads/tips-and-tricks-for-posting-the-coli-megathread.984734/post-52211196

1/12
@ai_for_success
China is on

ByteDance drops another banger AI paper on AI video!

- Goku : Flow-based video generative foundation model.
- Goku+ : Video ads foundation model - 100x lower cost than traditional ads methods.

These are insanely good!

Here are 10 incredible examples and the research paper link

https://video.twimg.com/ext_tw_video/1888819973256290304/pu/vid/avc1/1822x1080/BaJAjmuJr-bKlaki.mp4

2/12
@ai_for_success
2. Goku+: Turn Product Image To Video Clip

https://video.twimg.com/ext_tw_video/1888820045868134400/pu/vid/avc1/720x1280/vZIiiEj-LnlFSSZF.mp4

3/12
@ai_for_success
3. Goku+: Advertising Scenario

https://video.twimg.com/ext_tw_video/1888820133998817280/pu/vid/avc1/1822x1080/PYLXAIDn9NHXfkzP.mp4

4/12
@ai_for_success
4. Goku+: Advertising Scenario

https://video.twimg.com/ext_tw_video/1888820209831858176/pu/vid/avc1/1822x1080/VOdBzY4N2CclUCAA.mp4

5/12
@ai_for_success
Research Paper:

[Quoted tweet]
discuss: huggingface.co/papers/2502.0…
[media=twitter]1888811723047764367[/media]

6/12
@ai_for_success
5. Goku+: Product and Human Interaction

https://video.twimg.com/ext_tw_video/1888820329092702208/pu/vid/avc1/720x1280/WaMAqmDItiUAyUqz.mp4

7/12
@ai_for_success
6. Goku+: Create Marketing Avatar from Text

https://video.twimg.com/ext_tw_video/1888820394280538112/pu/vid/avc1/720x1280/GPWqPNyyF8hWKHNj.mp4

8/12
@ai_for_success
7. Goku text to Video : A capybara relaxes in a wooden barrel filled with steaming hot spring water, its serene gaze adding tranquility to the scene. Perched atop its head is a vibrant orange, adding a playful contrast to its soft brown fur.

https://video.twimg.com/ext_tw_video/1888820471480881152/pu/vid/avc1/1280x720/1HIvZeDcBtU1sxYY.mp4

9/12
@ai_for_success
8. Goku text to Video : Two women are sitting at a table in a room with wooden walls and a plant in the background. Both women look to the right and talk, with surprised expressions.

https://video.twimg.com/ext_tw_video/1888820536832372736/pu/vid/avc1/1280x720/AM53A5h9wqI2x2YR.mp4

10/12
@ai_for_success
9. Goku+ : Product and Human Interaction

https://video.twimg.com/ext_tw_video/1888820758794969088/pu/vid/avc1/720x1280/PZZ5DsEVYYC-7j9z.mp4

11/12
@ai_for_success
10. Goku+ : Advertising Scenario

https://video.twimg.com/ext_tw_video/1888820879653765120/pu/vid/avc1/1822x1080/rNkGImZ4FkRtiv1n.mp4

12/12
@ai_for_success
If you like this post follow me for more AI news @ai_for_success and Join my newsletter "AI Compass" for free and get all the latest AI News in your inbox.
AI Compass

To post tweets in this format, more info here: https://www.thecoli.com/threads/tips-and-tricks-for-posting-the-coli-megathread.984734/post-52211196

1/17
@CodeByPoonam
RIP Sora

China just dropped another open-source model: Goku, their Video Generator

13 wild examples so far (Don't miss the 5th one)

[Quoted tweet]
RIP Sora

China just dropped another open-source model: Goku, their Video Generator

13 wild examples so far (Don't miss the 5th one)
[media=twitter]1888900102313709699[/media]

To post tweets in this format, more info here: https://www.thecoli.com/threads/tips-and-tricks-for-posting-the-coli-megathread.984734/post-52211196

The A.I Megathread (LLM , GPT , Development)

Veteran

Pro

Veteran

Veteran

Hugging Face researchers aim to build an ‘open’ version of OpenAI’s deep research tool​

Veteran

Google launches new AI models and brings ‘thinking’ to Gemini​

Veteran

Researchers created an open rival to OpenAI’s o1 ‘reasoning’ model for under $50​

Veteran

Naver-backed Cinamon wants to make 3D video animation easier using AI​

Veteran

Meta says it may stop development of AI systems it deems too risky​

Veteran

Deepfake videos are getting shockingly good​

Veteran

Veteran

Veteran

Veteran

Veteran

Veteran

Hugging Face researchers aim to build an ‘open’ version of OpenAI’s deep research tool

Google launches new AI models and brings ‘thinking’ to Gemini

Researchers created an open rival to OpenAI’s o1 ‘reasoning’ model for under $50

Naver-backed Cinamon wants to make 3D video animation easier using AI

Meta says it may stop development of AI systems it deems too risky

Deepfake videos are getting shockingly good