Introducing Claude 2.1
Anthropic is an AI safety and research company that's working to build reliable, interpretable, and steerable AI systems.
www.anthropic.com
Large Language Models News & Discussions
- Thread starter Macallik86
- Start date
www.anthropic.com
Introducing Stable Video Diffusion — Stability AI
Stable Video Diffusion is a proud addition to our diverse range of open-source models. Spanning across modalities including image, language, audio, 3D, and code, our portfolio is a testament to Stability AI’s dedication to amplifying human intelligence.
bit.ly
About
Video-LLaVA: Learning United Visual Representation by Alignment Before Projectionarxiv.org/pdf/2311.10122.pdf
Video-LLaVA: Learning United Visual Representation by Alignment Before Projection
If you like our project, please give us a star 
on GitHub for latest update.
I also have other video-language projects that may interest you 
.
News
- [2023.11.20]
Demo and code are available now! Welcome to watch 
this repository for the latest updates.
Highlights
Video-LLaVA exhibits remarkable interactive capabilities between images and videos, despite the absence of image-video pairs in the dataset. Simple baseline, learning united visual representation by alignment before projection
- With the binding of unified visual representations to the language feature space, we enable an LLM to perform visual reasoning capabilities on both images and videos simultaneously.
High performance, complementary learning with video and image
- Extensive experiments demonstrate the complementarity of modalities, showcasing significant superiority when compared to models specifically designed for either images or videos.
Man. Seeing who is on the new board of OpenAI does not give me faith that we will be long for this world as a society.
A former CEO of Salesforce ? The board should have an ethical background, not a proven track record of being capitalistic. The foxes are officially in the hen house. I wonder if this is how people felt when Glass-Steagall was repealed.
Edit: fukking aye, just realized that Larry Summer who played a big part in repealing Glass-Steagall, is one of the other new board members
/cdn.vox-cdn.com/uploads/chorus_asset/file/25098718/90.jpeg)
Amazon’s offering free courses on generative AI
Could an Amazon AI skillset get you an AI job?
www.theverge.com
From the company that brought you AWS certification comes a new ‘AI Ready’ education track to help train aspiring professionals on Amazon’s AI tech
AWS Educate - Cloud Services for Education- AWS
AWS Educate provides students and educators with no-cost access to cloud computing skills training developed by AWS experts, plus career pathways for students.
aws.amazon.com
Self-paced digital training on AWS - AWS Skill Builder
Your learning center to build in-demand cloud skills.
explore.skillbuilder.aws
/cloudfront-us-east-2.images.arcpublishing.com/reuters/WRAGM7C3ONP4BO5NMF23ZGV2SA.jpg)
Exclusive: Sam Altman's ouster at OpenAI was precipitated by letter to board about AI breakthrough -sources
Ahead of OpenAI CEO Sam Altman’s four days in exile, several staff researchers sent the board of directors a letter warning of a powerful artificial intelligence discovery that they said could threaten humanity, two people familiar with the matter told Reuters.
www.reuters.com
That was quick.
Naming it Q Star is bout to have the MAGA crowd foaming at the mouth btw
Computer Science > Computation and Language
[Submitted on 15 Nov 2023 (v1), last revised 21 Nov 2023 (this version, v2)]Exponentially Faster Language Modelling
Peter Belcak, Roger Wattenhofer| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Neural and Evolutionary Computing (cs.NE) |
| Cite as: | arXiv:2311.10770 [cs.CL] |
| (or arXiv:2311.10770v2 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2311.10770 Focus to learn more |
Submission history
From: Peter Belcak [view email][v1] Wed, 15 Nov 2023 18:42:50 UTC (35 KB)
I’m excited to introduce RAGs - create a “ChatGPT over your data” purely with natural language 
Setup and use a full RAG pipeline over files/web urls in three simple steps:
RAG Builder: Describe your RAG pipeline in natural language (data, parameters like top-k, system prompt), and let the builder agent build this for you.
View Config: View the generated RAG configuration and make edits to it if you want to update the agent.
RAG Agent: Ask the agent anything and it’ll answer the question!
Of course, this is heavily inspired by GPTs from @OpenAI. The goal here was to build a similar natural language interface but tailored around RAG - retrieval parameters, prompts, data sources, all through conversation. Have an agent help you setup QA over your data instead of you writing code.
We built the app with @llama_index + @streamlit. It currently supports the following:
- LLMs: Supports our 20+ LLMs for builder agent (uses OpenAI by default) and currently supports OpenAI for generated agent (not too hard to extend).
- Embeddings: Supports OpenAI ada embeddings and @huggingface models.
There’s so many things we want to do. In the meantime if you have feedback/suggestions let us know!
Check out the attached video, blog post, and repo.
Blog: https://blog.llamaindex.ai/introduc...hatgpt-experience-over-your-data-2b9d140769b1
Repo: github.com/run-llama/rags
Setup and use a full RAG pipeline over files/web urls in three simple steps:
RAG Builder: Describe your RAG pipeline in natural language (data, parameters like top-k, system prompt), and let the builder agent build this for you. View Config: View the generated RAG configuration and make edits to it if you want to update the agent. RAG Agent: Ask the agent anything and it’ll answer the question!Of course, this is heavily inspired by GPTs from @OpenAI. The goal here was to build a similar natural language interface but tailored around RAG - retrieval parameters, prompts, data sources, all through conversation. Have an agent help you setup QA over your data instead of you writing code.
We built the app with @llama_index + @streamlit. It currently supports the following:
- LLMs: Supports our 20+ LLMs for builder agent (uses OpenAI by default) and currently supports OpenAI for generated agent (not too hard to extend).
- Embeddings: Supports OpenAI ada embeddings and @huggingface models.
There’s so many things we want to do. In the meantime if you have feedback/suggestions let us know!
Check out the attached video, blog post, and repo.
Blog: https://blog.llamaindex.ai/introduc...hatgpt-experience-over-your-data-2b9d140769b1
Repo: github.com/run-llama/rags
Google's Bard chatbot can now watch YouTube videos and provide information about them -
Microsoft recently retired the Bing Chat brand name for its generative AI chatbot, opting for the name Pilot. However, Google has maintained
www.fikrikadim.com
Google’s Bard chatbot can now watch YouTube videos and provide information about them
Google's Bard chatbot can now watch and provide information on YouTube video materialby FİKRİKADİM
November 22, 2023
2 mins read
Microsoft recently retired the Bing Chat brand name for its generative AI chatbot, opting for the name Pilot. However, Google has maintained the Bard name for its competing chatbot, at least for the time being. In the past week, Google introduced several new features to Bard, including the ability to engage in discussions about specific content from YouTube videos.
This new feature is an update to Google’s Bard Extensions launched in September, enabling Bard to integrate with various Google apps and services, including YouTube, to extract information for conversations. The latest update brings additional functionalities tailored specifically to YouTube.
In the official Bard update page, It states:
Alongside these YouTube-related enhancements, Google added other features to Bard last week. These include incorporating new “step-by-step explanations” for solving mathematical equations. Another noteworthy addition is Bard’s capacity to generate charts, either from user-inputted data or tables created during a conversation.
Moreover, Google reduced the minimum age requirement for Bard usage to 13 years old in certain countries. Notably, the company announced its intention to take action against two cybercriminal groups exploiting the Bard name for fraudulent activities. These actions include misleading individuals into downloading Bard under the false pretense of malware distribution.
İlgili
https://www.fikrikadim.com/2023/04/...tedposts_origin=81348&relatedposts_position=0Computer Science > Artificial Intelligence
[Submitted on 14 Nov 2023]LLMs cannot find reasoning errors, but can correct them!
Gladys Tyen, Hassan Mansoor, Peter Chen, Tony Mak, Victor Cărbune| Subjects: | Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG) |
| Cite as: | arXiv:2311.08516 [cs.AI] |
| (or arXiv:2311.08516v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2311.08516 Focus to learn more |
Submission history
From: Gladys Tyen [view email][v1] Tue, 14 Nov 2023 20:12:38 UTC (7,191 KB)
GitHub - WHGTyen/BIG-Bench-Mistake: A dataset of LLM-generated chain-of-thought steps annotated with mistake location.
A dataset of LLM-generated chain-of-thought steps annotated with mistake location. - GitHub - WHGTyen/BIG-Bench-Mistake: A dataset of LLM-generated chain-of-thought steps annotated with mistake lo...
github.com
BIG-Bench-Hard/cot-prompts at main · suzgunmirac/BIG-Bench-Hard
Challenging BIG-Bench Tasks and Whether Chain-of-Thought Can Solve Them - suzgunmirac/BIG-Bench-Hard
github.com
partial AI summary:
ANSWER | PHIND V8 MODEL
The paper you're referring to is about Large Language Models (LLMs) and their ability to self-correct their outputs. The authors, Gladys Tyen, Hassan Mansoor, Peter Chen, Tony Mak, and Victor Carbune, have broken down the self-correction process into two components: mistake finding and output correction.Here are the key highlights:
- Mistake Finding: The authors have released a dataset called BIG-Bench Mistake, which contains logical mistakes in Chain-of-Thought reasoning traces. They found that LLMs generally struggle with finding these logical mistakes.
- Output Correction: The authors propose a backtracking method for output correction. This method provides large improvements when given information on the mistake location. They view backtracking as a lightweight alternative to reinforcement learning methods and show that it remains effective with a reward model at 60-70% accuracy.
- Benchmark Results: The authors tested the reasoning capabilities of state-of-the-art LLMs using their dataset. They found that LLMs struggle with mistake finding, even for objective, unambiguous cases. This suggests that LLMs' inability to find mistakes is a main contributing factor to their inability to self-correct reasoning errors.
- Backtracking as Verbal Reinforcement Learning: The authors view backtracking as a form of "verbal reinforcement learning". They propose that backtracking can be used with a trained classifier as a reward model, and demonstrate the effectiveness of backtracking at various reward model accuracies.
- BIG-Bench Mistake Dataset: The authors created the BIG-Bench Mistake dataset, which is the first of its kind that goes beyond problems in mathematics. The dataset consists of 2186 sets of CoT-style traces, each annotated with the location of the first logical error. The authors used this dataset to test the reasoning capabilities of state-of-the-art LLMs.
- Annotation: The authors used both human and automatic annotation for the BIG-Bench Mistake dataset. They found that the inter-rater reliability for the human-annotated tasks was high, suggesting the reliability of the dataset.
- Prompting Methods: The authors compared three different prompting methods: Direct trace-level prompting, Direct step-level prompting, and CoT step-level prompting. They found that the accuracy on traces with no mistakes goes down considerably from direct, trace-level prompting to CoT step-level prompting.
- Few-shot Prompting for Mistake Location: The authors investigated whether their prompting methods can reliably determine the correctness of a trace rather than the mistake location. They found that using mistake finding as a strategy for determining whether an answer is correct or not can be effective.
