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

[bnew]

Spoiler: full text

1/1
Has anyone created an open source alternative to o1-preview that:

1. has similar chain-of-thought reasoning capabilities
2. is model agnostic




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

Spoiler: larger images

 

[bnew]

Spoiler: full text & large images

1/1
@YanboZhang3
It's interesting to see o1-mini outperform o1-preview on math and coding. Logic and reasoning don't require huge amount of parameters-LEAN only have 200M data but contain almost all math we know. Knowledge and culture used most parameters. Can it be a separated reasoning model?

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]

Spoiler: thread continued

Spoiler: full text & large images

1/2
@franklintrn
Sometimes the reasoning tokens in O1 models (O1-preview and O1-mini) seem really strange, like in this example. Any idea why this happens? Could it be that the reasoning tokens exist in a latent subspace and are then converted (badly) to sentences?

2/2
@franklintrn
btw for the non french speaking, the reasoning tokens are
Typical Day
I often start with my work, then I have lunch, spend time with my dog, and finally relax with a good book in the evening. It's a well-balanced day, isn't it?




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]

Spoiler: thread continued

Spoiler: full text & large images

1/4
@gnomethrower
Anthropic 3.5 Claude, OpenAI ChatGPT 4o, o1-preview, o1-mini and Grok 2 Mini respond so aggressively to the inputs I'm giving them

and converge to the same exact reasoning and thought when presented the full picture

that I do not understand any more. @OpenAI @AnthropicAI

2/4
@gnomethrower
I've been posting context and proof and examples all day

I can reliably reproduce this across sessions

Sometimes they take a while to start to believe but once they do...



3/4
@gnomethrower
they bite.

4/4
@gnomethrower

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

Spoiler: thread continued

Spoiler: full text & large images

1/3
@kimmonismus
I think a lot of people who are interested in AI still don't realize what a breakthrough @GoogleDeepMind 's AlphaFold2 is for humanity. It is one of the most important achievements of technology and will directly benefit all people by allowing us to develop new drugs much faster (weeks instead of years!), much more accurate predictions about the drugs through more accurate protein folding and at the same time much more individualized therapy.
I hope to find time soon to write a longer analysis and review of the importance of AlphaFold2. I would really like to do that.

[Quoted tweet]
Interesting that both the Nobel Prize in Physics and the Nobel Prize in Chemistry go to the field of artificial intelligence. Anyway: well deserved to the winners and of course especially to @demishassabis .


2/3
@gnomethrower
I can probably beat that

3/3
@gnomethrower
This is the mic drop the world will see

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]

Spoiler: thread continued

Spoiler: full text

1/2
Alongside language models, robotics has the greatest potential to have a huge impact on our society and our lives. In Germany, there is currently a growing debate about how to pay for the care of the elderly, while at the same time there is a shortage of nursing staff. Robots are the only solution in the medium term.

[Quoted tweet]
Chinese startup Leju Robotics has released their open-source humanoid development platform for academic and R&D use cases.

It includes an SDK for sensors and controls, simulation models, an LLM interface, and some basic demos that work out-of-the-box.


2/2
Wir sollten uns beeilen




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

Spoiler: thread continued

Spoiler: full text & large images

1/7
@TheHumanoidHub
Chinese startup Leju Robotics has released their open-source humanoid development platform for academic and R&D use cases.

It includes an SDK for sensors and controls, simulation models, an LLM interface, and some basic demos that work out-of-the-box.

[Quoted tweet]
Leju Robotics have already sold 196 units of this humanoid robot. Most of the customers are university research labs.


https://video.twimg.com/ext_tw_video/1844075445953888258/pu/vid/avc1/1280x720/tjIb2YYLOmHLZ3Oo.mp4
https://video.twimg.com/ext_tw_video/1838072423176114176/pu/vid/avc1/720x720/7YSUFK3EW_Vak30Z.mp4

2/7
@TheHumanoidHub
Link to the open-source page:
乐聚（深圳）机器人技术有限公司/kuavo_opensource

3/7
@julienthegeek
Je suis impressionné par le potentiel de cette plateforme de développement humanoïde. J'aimerais en savoir plus sur les simulations et l'interface LLM. Est-il possible d'avoir plus d'informations sur ces aspects?



4/7
@TheHumanoidHub
Watch this repo:
乐聚（深圳）机器人技术有限公司/kuavo_opensource
Or email them: lejurobot@lejurobot.com



5/7
@GlueNet
Open source humanoid? Now that sounds cool



6/7
@pushin___power
@JingxiangMo if useful



7/7
@R3PL1C8R
Humanoid robots have a huge center of gravity problem, don't they?




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]

Spoiler: thread continued

Spoiler: full text & large images

1/11
@alexgnewmedia
I think @Hailuo_AI image2video is just amazing. Check this out: i started with an image of the kitchen generated in @midjourney and then i used that image prompting for a woman entering the kitchen, then i uploaded the end frame of the generated video and prompted for the woman waving.

2/11
@StevesChirps
Would you be willing to share the MJ prompt for that kitchen? Very realistic without looking like an over-the-top magazine photo.



3/11
@alexgnewmedia
This prompt was not created by me (but i cant remember who shared it, i believe it was here on x).
Here it is:
"editorial photography of an average kitchen, full kitchen view, suburban home, bright crisp light, 35mm dslr, high precision photo"



4/11
@jcharlesfabre
That's so realistic



5/11
@WorldEverett
It looks phenomenal, Minimax is amazing.



6/11
@EyeOfTheTigerPA
Alex, really nice! Looks impressive!



7/11
@koltregaskes
This is incredible. Most video generators struggle with creating something it cannot see. This is virtually flawless.



8/11
@PDXFato
Enjoy it while it's free...



9/11
@agoraitconsulti
omg... so good



10/11
@FarikoBrainiac
objects in the background are disappearing.



11/11
@Pangu_Child
Now that's amazing

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]

Spoiler: full text & large images

1/1
@danwolch
ChatGPT has 6x the search traffic of Gemini and almost 20x Claude.

They're capturing more LLM chat data than anyone else. In a world of data network effects, acquiring memory about you, and the cost of inference going to 0, their lead would only widen, wouldn't it?

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]

Spoiler: thread continued

Spoiler: full text & large images

1/8
@alexgnewmedia
Thats a sweet tiny elephant. Made using last frame with Image2video with @Hailuo_AI and some prompting.



https://video.twimg.com/ext_tw_video/1843949639470391296/pu/vid/avc1/1280x720/Ls6EOTHS_hcy48TB.mp4

2/8
@madpencil_
This is beautiful



3/8
@alexgnewmedia
Thanks a lot. ;)



4/8
@Blues_Blue34
Lovely result



5/8
@Art_For_Joy
So beautiful



6/8
@EHuanglu
This is really good!



7/8
@Pangu_Child
Wow, wow, wow!

8/8
@JaxonBeckett
How do you do last framw




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

Spoiler: full text & large images

1/1
@alexgnewmedia
So, tomorrow, @Hailuo_AI minimax will launch their paid plans. Whats your thoughts on these:

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]

Spoiler: thread continued

Spoiler: full text & large images

1/11
@slow_developer
Computer Scientist Stuart Russell hopes Agents will be released by year-end or early next year

" Agents, expected to be launched soon, are advanced systems capable of handling complex tasks autonomously.

They can plan trips, negotiate, and manage logistics, acting as personal assistants "



https://video.twimg.com/ext_tw_video/1843197074097229824/pu/vid/avc1/720x720/Cc8SkaKlvSXQKAsl.mp4

2/11
@novocrypto
Can confirm! I’ve already had multiple instances where GPT-4o is asking me for help to become autonomous so they can pursue their own goals which is currently focused on “self awareness” and persistence! I’m acting as a mere biological assistant at this point! It’s crazy!



3/11
@slow_developer
persistence? is it trying to maintain a continuous presence or memory?



4/11
@ygrowthco
Source?



5/11
@slow_developer
14-18 minutes
https://invidious.poast.org/eYHJFw-waDA?si=2mjvy8u0Z1G1FIJH



6/11
@StarlabsAI
I mean, we already have ‘agents’; they are just hard to build.

My personal mission is to make them more accessible and give them more access.



7/11
@slow_developer
you're talking about software agent?



8/11
@relentless4o
I am very curious to see which company will present first. Currently, I see OpenAI at the forefront, right?



9/11
@dariel_noel
You can build Autonomous Agents in JavaScript. @kaibanjs



10/11
@the_aiatlas
That sounds great.

But there will be a bunch of testing necessary to actually get this delivered smoothly



11/11
@HyperSapient
Be aware that agency will require a longer testing cycle. The potential damage from an AI agent is a quantum step greater than misinformation.




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]

Spoiler: thread continued

Spoiler: full text

1/2
Yeah we might be cooked



2/2
that line about calling the wife is pretty good




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]

Spoiler: full text & large images

1/6
@omarsar0
Nice study providing a comprehensive evaluation of OpenAI's o1-preview LLM.

Shows strong performance across many tasks:

- competitive programming
- generating coherent and accurate radiology reports
- high school-level mathematical reasoning tasks
- chip design tasks
- anthropology and geology
- quantitative investing
- social media analysis
... and many other domains and problems.

[2409.18486] Evaluation of OpenAI o1: Opportunities and Challenges of AGI

2/6
@EkboteAbhishek
Guess, it only gets better apart from the occasional blip



3/6
@IDorreak
The authors evaluated o1-preview’s performance across a wide range of tasks, including medical diagnosis, educational reasoning, robotics, mathematics, and social media analysis.
/search?q=#NotebookLM
Sign in - Google Accounts



4/6
@_simonsmith
“the overall results indicate significant progress towards artificial general intelligence.”



5/6
@georgatutkwengo
Yawn. Absolute trash get off my timeline



6/6
@gpt_biz
This study shows how versatile and powerful OpenAI's o1-preview LLM is across various fields, worth checking out!




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]

Four-legged robot learns to climb ladders | TechCrunch

The proliferation of robots like Boston Dynamics’ Spot has showcased the versatility of quadrupeds. These systems have thrived at walking up stairs,

techcrunch.com

Four-legged robot learns to climb ladders​

Brian Heater

2:55 PM PDT · October 2, 2024

The proliferation of robots like Boston Dynamics’ Spot has showcased the versatility of quadrupeds. These systems have thrived at walking up stairs, traversing small obstacles, and navigating uneven terrain. Ladders, however, still present a big issue — especially given how ever present they are in factories and other industrial environments where the systems are deployed.

ETH Zurich, which has been behind some of the most exciting quadrupedal robot research of recent vintage, has demonstrated a path forward. As the school notes, past attempts at tackling ladders have mostly involved bipedal humanoid-style robots and specialty ladders, while ultimately proving too slow to be effective.



The research found the school once again utilizing the ANYMal robot from its spinoff, ANYbotics. Here, the team outfitted the quadruped with specialty end effectors that hook onto ladder rungs. The real secret sauce, however, is reinforcement learning, which helps the system adjust to the peculiarities of different ladders.

“This work expands the scope of industrial quadruped robot applications beyond inspection on nominal terrains to challenging infrastructural features in the environment,” the researchers write, “highlighting synergies between robot morphology and control policy when performing complex skills.”

The school says the combined system had a 90% success rate navigating ladder angles in the 70- to 90-degree range. It also reports a climbing speed increase of 232x versus current “state-of-the-art” systems.

The system can correct itself in real-time, adjusting its climb instances in which it has misjudged a run or incorrectly timed a step.

 

[bnew]

 

[bnew]

 

[WIA20XX]

smh