AI/ML

AI/ML

2490 bookmarks
Custom sorting
Mojo šŸ”„ ā€” a new programming language for AI developers (first look)
Mojo šŸ”„ ā€” a new programming language for AI developers (first look)
Yesterday Chris Lattner announced a new programming language called Mojo, which builds on top of Python and is designed specifically for AI applications. They claim itā€™s up to 35000 times faster than Python. So letā€™s take a quick first look at it. Website: https://www.modular.com/mojo Docs: https://docs.modular.com/mojo/ Keynote: https://youtu.be/-3Kf2ZZU-dg Get my Free NumPy Handbook: https://www.python-engineer.com/numpybook āœ… Write cleaner code with Sourcery, instant refactoring suggestions in VS Code & PyCharm: https://sourcery.ai/?utm_source=youtube&utm_campaign=pythonengineer * ā­ Join Our Discord : https://discord.gg/FHMg9tKFSN šŸ““ ML Notebooks available on Patreon: https://www.patreon.com/patrickloeber If you enjoyed this video, please subscribe to the channel: ā–¶ļø : https://www.youtube.com/channel/UCbXgNpp0jedKWcQiULLbDTA?sub_confirmation=1 ~~~~~~~~~~~~~~~ CONNECT ~~~~~~~~~~~~~~~ šŸ–„ļø Website: https://www.python-engineer.com šŸ¦ Twitter - https://twitter.com/patloeber āœ‰ļø Newsletter - https://www.python-engineer.com/newsletter šŸ“ø Instagram - https://www.instagram.com/patloeber šŸ¦¾ Discord: https://discord.gg/FHMg9tKFSN ā–¶ļø Subscribe: https://www.youtube.com/channel/UCbXgNpp0jedKWcQiULLbDTA?sub_confirmation=1 ~~~~~~~~~~~~~~ SUPPORT ME ~~~~~~~~~~~~~~ šŸ…æ Patreon - https://www.patreon.com/patrickloeber #Python Timeline: 00:00 - Intro & motivation 01:23 - Features 04:48 - How to try Mojo ---------------------------------------------------------------------------------------------------------- * This is an affiliate link. By clicking on it you will not have any additional costs, instead you will support me and my project. Thank you so much for the support! šŸ™
Ā·youtube.comĀ·
Mojo šŸ”„ ā€” a new programming language for AI developers (first look)
Inflection AI, Startup From Ex-DeepMind Leaders, Launches Pi ā€” A Chattier Chatbot
Inflection AI, Startup From Ex-DeepMind Leaders, Launches Pi ā€” A Chattier Chatbot
Mustafa Suleyman, CEO of the year-old startup thatā€™s already raised $225 million and claims to run one of the worldā€™s largest language models, sees his dialog-based chatbot as a key step toward a true AI-based personal assistant.
Ā·forbes.comĀ·
Inflection AI, Startup From Ex-DeepMind Leaders, Launches Pi ā€” A Chattier Chatbot
Langchain gpt-3.5-turbo models reads files - problem
Langchain gpt-3.5-turbo models reads files - problem
I am making really simple (and for fun) LangChain project. A model can read PDF file and I can then ask him questions about specific PDF file. Everything works fine (this is working example) from P...
#text
Ā·stackoverflow.comĀ·
Langchain gpt-3.5-turbo models reads files - problem
Modal
Modal
Ā·modal.comĀ·
Modal
Stability AI releases StableVicuna, the AI Worldā€™s First Open Source RLHF LLM Chatbot ā€” Stability AI
Stability AI releases StableVicuna, the AI Worldā€™s First Open Source RLHF LLM Chatbot ā€” Stability AI
Experience the power of StableVicuna, the first large-scale open source chatbot trained via reinforced learning from human feedback (RHLF). With enhanced training and fine-tuning capabilities, StableVicuna offers advanced chatbot solutions to drive engagement and improve customer interactions. Try S
Ā·stability.aiĀ·
Stability AI releases StableVicuna, the AI Worldā€™s First Open Source RLHF LLM Chatbot ā€” Stability AI
OpenAI Threatens Popular GitHub Project With Lawsuit Over API Use
OpenAI Threatens Popular GitHub Project With Lawsuit Over API Use
GPT4Free uses other sites' connections to OpenAI.
A GitHub project called GPT4free (opens in new tab) allows you to get free access to the GPT4 and GPT3.5 models by funneling those queries through sites like You.com (opens in new tab), Quora (opens in new tab) and CoCalc (opens in new tab) and giving you back the answers. The project is GitHub's most popular new repo, getting 14,000 stars this week.
On the backend, GPT4Free is visiting various API urls that sites like You.com, an AI-powered search engine that employs OpenAI's GPT3.5 model for its answers, use for their own queries. For example, the main GPT4Free script hits the URL https://you.com/api/streamingSearch, feeds it various parameters, and then takes the JSON it returns and formats it. The GPT4Free repo also has scripts that grab data from other sites such as Quora, Forefront, and TheB. Any enterprising developer could use these simple scripts to make their own bot.
If those sites are relying on ad revenue from their sites to offset these API costs, they are losing money because of these queries.
Perhaps more importantly, Xtrekky noted, any of these sites could block external uses of their internal APIs with common security measures.
Xtrekky said that he has advised all the sites that wrote to him that they should secure their APIs, but none of them has done so. So, even if he takes the scripts down from his repo, any other developer could do the same thing.
#IP#legal
Ā·tomshardware.comĀ·
OpenAI Threatens Popular GitHub Project With Lawsuit Over API Use
MLC LLM
MLC LLM
From MLC, the team that gave us Web LLM and Web Stable Diffusion. "MLC LLM is a universal solution that allows any language model to be deployed natively on a ā€¦
#iOS
Ā·simonwillison.netĀ·
MLC LLM
Lawmakers propose banning AI from singlehandedly launching nuclear weapons
Lawmakers propose banning AI from singlehandedly launching nuclear weapons
Pentagon policy already requires a ā€œhuman in the loop.ā€
As announced earlier this week, Senator Edward Markey (D-MA) and Representatives Ted Lieu (D-CA), Don Beyer (D-VA), and Ken Buck (R-CO) have introduced the Block Nuclear Launch by Autonomous AI Act, which would ā€œprohibit the use of Federal funds to launch a nuclear weapon using an autonomous weapons system that is not subject to meaningful human control.ā€ The act would codify existing Pentagon rules for nuclear weapons, which, as of 2022, read thusly:
Its sponsors note that a 2021 National Security Commission on Artificial Intelligence report recommended affirming a ban on autonomous nuclear weapons launches, not only to prevent it from happening inside the US government but to spur similar commitments from China and Russia. Publicizing the bill calls attention to the potential dangers of current-generation autonomous artificial intelligence systems, a going concern in Congress and the tech world alike. And as indicated by the press release, it offers a chance to highlight the sponsorsā€™ other nuclear non-proliferation efforts ā€”Ā like a recent bill restricting the presidentā€™s power to unilaterally declare nuclear war.
Ā·theverge.comĀ·
Lawmakers propose banning AI from singlehandedly launching nuclear weapons
Has Larry Lessig Lost The Plot? Tells Supreme Court That AI Should Get Patents
Has Larry Lessig Lost The Plot? Tells Supreme Court That AI Should Get Patents
Larry Lessigā€™s views and thoughts on things like copyright law, internet freedom, and government corruption have been tremendously influential on myself and many others in the tech and tech policy ā€¦
Under US law, the issue is quite clear: you need an inventor and an inventor needs to be human, to get a patent. There are many good reasons for this: mainly because the entire point of the patent system is to create incentives to invent. An AI systemā€¦ doesnā€™t need that incentive. It just responds to inputs.
Iā€™m sorry, but what?!? Larry Lessig arguing that not giving out patents to AI ā€œjeopardizes billions in current and future investments?ā€ This is the same Lessig, after all, who created Creative Commons and spent years and multiple books explaining how locking up knowledge via intellectual monopolies was harmful
First of all, most inventions wouldnā€™t be protected by trade secrets anyway, because most things can be reverse engineered. Second, the value in inventing something is in bringing that product to market. Thatā€™s where you make the money, in selling the product, not the patent. Third, the point of the patent is to help the inventor recoup the capital expenditure in creating the invention in the first place. But, AI is cheap and can generate tons of ideas quickly. The capex is minimal.
If the makers of AI are concerned that they canā€™t make money without patents, just ask their AI to invent products they can sell in the market. They donā€™t need patents. They need products.
#legal#IP
Ā·techdirt.comĀ·
Has Larry Lessig Lost The Plot? Tells Supreme Court That AI Should Get Patents
microsoft/table-transformer: Table Transformer (TATR) is a deep learning model for extracting tables from unstructured documents (PDFs and images). This is also the official repository for the PubTables-1M dataset and GriTS evaluation metric.
microsoft/table-transformer: Table Transformer (TATR) is a deep learning model for extracting tables from unstructured documents (PDFs and images). This is also the official repository for the PubTables-1M dataset and GriTS evaluation metric.
Table Transformer (TATR) is a deep learning model for extracting tables from unstructured documents (PDFs and images). This is also the official repository for the PubTables-1M dataset and GriTS ev...
#text#text sanitization
Ā·github.comĀ·
microsoft/table-transformer: Table Transformer (TATR) is a deep learning model for extracting tables from unstructured documents (PDFs and images). This is also the official repository for the PubTables-1M dataset and GriTS evaluation metric.
Knowledge Retrieval Architecture for LLMā€™s (2023)
Knowledge Retrieval Architecture for LLMā€™s (2023)
In this guide, I will share the standard architecture for data-informed language model applications and explain forthcoming improvements in knowledge retrieval.
At a high level, there are two primary methods for referencing specific data:Insert data as context in the model prompt, and direct the response to utilize that informationFine-tune a model, by providing hundreds or thousands of prompt <> completion pairs
The design above goes by various names, most commonly "retrieval-augmented generation" or "RETRO"
Retrieval-augmented generation a) retrieves relevant data from outside of the language model (non-parametric) and b) augments the data with context in the prompt to the LLM. The architecture cleanly routes around most of the limitations of fine-tuning and context-only approaches.
Ā·mattboegner.comĀ·
Knowledge Retrieval Architecture for LLMā€™s (2023)
AI is taking the jobs of Kenyans who write essays for U.S. college students
AI is taking the jobs of Kenyans who write essays for U.S. college students
Ghostwriters say the meteoric rise of ChatGPT has coincided with a drop in income.
In January 2023, online learning platform Study surveyed more than 1,000 American students and over 100 educators. More than 89% of the students said they had used ChatGPT for help with a homework assignment. Nearly half admitted to using ChatGPT for an at-home test or quiz, 53% had used it to write an essay, and 22% had used it for outlining one.
While 17 states in the U.S. have banned contract cheating, it has not been a problem for freelancers in Kenya
Christopher Kanan, an associate professor in the department of computer science at the University of Rochester, has started giving in-person, in-class quizzes due to the popularity of ChatGPT
Others, like Ethan Mollick, an associate professor at the University of Pennsylvaniaā€™s Wharton School, have chosen to take a more open approach to ChatGPT in class. ā€œThe truth is, I probably couldnā€™t have stopped them even if I didnā€™t require it,ā€ he told NPR.
#culture#economics
Ā·restofworld.orgĀ·
AI is taking the jobs of Kenyans who write essays for U.S. college students
The Dual LLM pattern for building AI assistants that can resist prompt injection
The Dual LLM pattern for building AI assistants that can resist prompt injection
I really want an AI assistant: a Large Language Model powered chatbot that can answer questions and perform actions for me based on access to my private data and tools. ā€¦
Confused deputy attacks Confused deputy is a term of art in information security. Wikipedia defines it like this: In information security, a confused deputy is a computer program that is tricked by another program (with fewer privileges or less rights) into misusing its authority on the system. It is a specific type of privilege escalation.
Language model applications work by mixing together trusted and untrusted data sources
For example, if the LLM generates instructions to send or delete an email the wrapping UI layer should trigger a prompt to the user asking for approval to carry out that action.
More to the point, it will inevitably suffer from dialog fatigue: users will learn to click ā€œOKā€ to everything as fast as possible, so as a security measure itā€™s likely to catastrophically fail.
Data exfiltration attacks Wikipedia definition: Data exfiltration occurs when malware and/or a malicious actor carries out an unauthorized data transfer from a computer. It is also commonly called data extrusion or data exportation. Data exfiltration is also considered a form of data theft.
Even if an AI agent canā€™t make its own HTTP calls directly, there are still exfiltration vectors we need to lock down.
Locking down an LLM Weā€™ve established that processing untrusted input using an LLM is fraught with danger. If an LLM is going to be exposed to untrusted contentā€”content that could have been influenced by an outside attacker, via emails or web pages or any other form of untrusted inputā€”it needs to follow these rules: No ability to execute additional actions that could be abused And if it might ever mix untrusted content with private data that could be the target of an exfiltration attack: Only call APIs that can be trusted not to leak data No generating outbound links, and no generating outbound images This is an extremely limiting set of rules when trying to build an AI assistant. It would appear to rule out most of the things we want to build!
For any output that could itself host a further injection attack, we need to take a different approach. Instead of forwarding the text as-is, we can instead work with unique tokens that represent that potentially tainted content.
#security#prompt
Ā·simonwillison.netĀ·
The Dual LLM pattern for building AI assistants that can resist prompt injection
deep-floyd/IF
deep-floyd/IF
Contribute to deep-floyd/IF development by creating an account on GitHub.
Ā·github.comĀ·
deep-floyd/IF
A.I. and Stochastic Parrots | FACTUALLY with Emily Bender and Timnit Gebru
A.I. and Stochastic Parrots | FACTUALLY with Emily Bender and Timnit Gebru
SUBSCRIBE TO FACTUALLY: https://link.chtbl.com/nD2_iAuV SUPPORT THE SHOW ON PATREON: http://patreon.com/adamconover So-called ā€œartificial intelligenceā€ is one of the most divisive topics of the year, with even those who understand it in total disagreement about its potential impacts. This week, A.I. reseachers and authors of the famous paper ā€œOn the Dangers of Stochastic Parrots,ā€ Emily Bender and Timnit Gebru, join Adam to discuss what everyone gets wrong about A.I.
Ā·youtube.comĀ·
A.I. and Stochastic Parrots | FACTUALLY with Emily Bender and Timnit Gebru
Investigating MiniGPT-4 - The Secret behind GPT-V?
Investigating MiniGPT-4 - The Secret behind GPT-V?
Project: https://minigpt-4.github.io/Demo: https://c8de8ff74b6a6c6a9b.gradio.live/In this video I look at the project MiniGPT-4: Enhancing Vision-language Un...
Ā·youtube.comĀ·
Investigating MiniGPT-4 - The Secret behind GPT-V?
Minigpt-4
Minigpt-4
Ā·minigpt-4.github.ioĀ·
Minigpt-4