Last Week in AI Skynet Today

Our 175th episode with a summary and discussion of last week's big AI news!
With hosts Andrey Kurenkov (https://twitter.com/andrey_kurenkov) and Jeremie Harris (https://twitter.com/jeremiecharris)
In this episode of Last Week in AI, hosts Andrey Kurenkov and Jeremy Harris explore recent AI advancements including OpenAI's release of GPT 4.0 Mini and Mistral’s open-source models, covering their impacts on affordability and performance. They delve into enterprise tools for compliance, text-to-video models like Hyper 1.5, and YouTube Music enhancements. The conversation further addresses AI research topics such as the benefits of numerous small expert models, novel benchmarking techniques, and advanced AI reasoning. Policy issues including U.S. export controls on AI technology to China and internal controversies at OpenAI are also discussed, alongside Elon Musk's supercomputer ambitions and OpenAI’s Prover-Verify Games initiative.
 
Read out our text newsletter and comment on the podcast at https://lastweekin.ai/
If you would like to become a sponsor for the newsletter, podcast, or both, please fill out this form.
Email us your questions and feedback at contact@lastweekinai.com and/or hello@gladstone.ai
 
Timestamps + links:
(00:00:00) AI Song Intro
(00:00:40) Intro / Banter
Tools & Apps(00:03:57) OpenAI unveils GPT-4o mini, a small AI model powering ChatGPT
(00:11:38) Meet Haiper 1.5, the new AI video generation model challenging Sora, Runway
(00:16:32) Anthropic releases Claude app for Android
(00:18:59) Google Vids is available to test out Gemini AI-created video presentations
(00:20:27) YouTube Music sound search rolling out, AI ‘conversational radio’ in testing 

Applications & Business(00:23:30) OpenAI working on new reasoning technology under code name ‘Strawberry’
(00:30:45) Inside Elon Musk’s Mad Dash To Build A Giant xAI Supercomputer In Memphis
(00:37:15) Apple, NVIDIA and Anthropic reportedly used YouTube transcripts without permission to train AI models
(00:41:05) After Tesla and OpenAI, Andrej Karpathy’s startup aims to apply AI assistants to education
(00:43:40) Menlo Ventures and Anthropic team up on a $100M AI fund

Projects & Open Source(00:46:27) Mistral releases Codestral Mamba for faster, longer code generation
(00:50:36) Mistral AI and NVIDIA Unveil Mistral NeMo 12B, a Cutting-Edge Enterprise AI Model
(00:52:51) Hugging Face Releases SmoLLM, a Series of Small Language Models, Beats Qwen2 and Phi 1.5
(00:56:11) Stable Diffusion 3 License Revamped Amid Blowback, Promising Better Model

Research & Advancements(01:01:49) FlashAttention-3 unleashes the power of H100 GPUs for LLMs
(01:06:38) Mixture of A Million Experts
(01:12:51) AutoBencher: Creating Salient, Novel, Difficult Datasets for Language Models
(01:18:23) SpreadsheetLLM: Encoding Spreadsheets for Large Language >Models

Policy & Safety(01:20:50) Prover-Verifier Games improve legibility of language model outputs
(01:28:05) Trump allies draft AI order to launch ‘Manhattan Projects’ for defense
(01:34:40) On scalable oversight with weak LLMs judging strong LLMs
(01:36:24) Google, Microsoft offer Nvidia chips to Chinese companies, the Information reports
(01:38:26) U.S. planning 'draconian' sanctions against China's semiconductor industry: Report
(01:48:47) OpenAI illegally barred staff from airing safety risks, whistleblowers say

(01:44:59) Outro + AI Song

Our 174rd episode with a summary and discussion of last week's big AI news!
With hosts Andrey Kurenkov (https://twitter.com/andrey_kurenkov) and Jeremie Harris (https://twitter.com/jeremiecharris)
In this episode of Last Week in AI, we delve into the latest advancements and challenges in the AI industry, highlighting new features from Figma and Quora, regulatory pressures on OpenAI, and significant investments in AI infrastructure. Key topics include AMD's acquisition of Silo AI, Elon Musk's GPU cluster plans for XAI, unique AI model training methods, and the nuances of AI copying and memory constraints. We discuss developments in AI's visual perception, real-time knowledge updates, and the need for transparency and regulation in AI content labeling and licensing.
See full episode notes here.
Read out our text newsletter and comment on the podcast at https://lastweekin.ai/
If you would like to become a sponsor for the newsletter, podcast, or both, please fill out this form.
Email us your questions and feedback at contact@lastweekinai.com and/or hello@gladstone.ai
 
Timestamps + links:
(00:00:00) Intro AI Song
(00:00:41) Pre News Banter
Tools & Apps(00:07:09) Odyssey Building 'Hollywood-Grade' AI Text-to-Video Model to Compete With Sora, Gen-3 Alpha
(00:10:28) Anthropic’s Claude adds a prompt playground to quickly improve your AI apps
(00:15:06) Figma pauses its new AI feature after Apple controversy
(00:18:30) Quora’s Poe now lets users create and share web apps
(00:20:54) Suno launches iPhone app — now you can make AI music on the go

Applications & Business(00:21:42) Groq unveils lightning-fast LLM engine; developer base rockets past 280K in 4 months
(00:27:03) Microsoft and Apple ditch OpenAI board seats amid regulatory scrutiny
(00:29:39) OpenAI and Arianna Huffington are working together on an ‘AI health coach’
(00:33:38) AI coding startup Magic seeks $1.5-billion valuation in new funding round, sources say
(00:37:01) Sequoia and Andreessen Horowitz Clash Over AI Chip Supplies Amid Gen AI Boom
(00:43:30) Elon Musk Reveals Plans To Make World’s “Most Powerful” 100,000 NVIDIA GPU AI Cluster
(00:46:25) AMD plans to acquire Silo AI in $665 million deal
(00:48:00) AI robotics startup raises US$300 million, including from Jeff Bezos
(00:52:11) Intel begins groundwork on Magdeburg chip fab despite 13 remaining regulatory and environmental objections

Research & Advancements(00:55:21) Learning to (Learn at Test Time): RNNs with Expressive Hidden States
(01:03:12) Data curation via joint example selection further accelerates multimodal learning
(01:09:11) CopyBench: Measuring Literal and Non-Literal Reproduction of Copyright-Protected Text in Language Model Generation
(01:13:25) Just read twice: closing the recall gap for recurrent language models
(01:15:25) CodeUpdateArena: Benchmarking Knowledge Editing on API Updates
(01:18:31) Composable Interventions for Language Models
(01:24:09) Mind-reading AI recreates what you're looking at with amazing accuracy

Policy & Safety(01:26:49) Covert Malicious Finetuning
(01:31:23) OpenAI’s week of security issues
(01:36:39) Here’s how OpenAI will determine how powerful its AI systems are
(01:39:56) Me, Myself and AI: The Situational Awareness Dataset for LLMs
(01:44:34) Exclusive: OpenAI partners with Los Alamos to study AI in the lab
(01:47:36) Judge dismisses coders’ DMCA claims against Microsoft, OpenAI and GitHub
(01:49:55) A former OpenAI safety employee said he quit because the company's leaders were 'building the Titanic' and wanted 'newer, shinier' things to sell

Synthetic Media & Art(01:52:46) Vimeo joins YouTube and TikTok in launching new AI content labels
(01:54:50) Tech Startup Aims to Help Media License Content for AI Training
(01:57:23) Etsy adds AI-generated item guidelines in new seller policy 
(01:59:44) Bumble users can now report profiles that use AI-generated photos

(02:02:05) Outro + AI Song

Our 173rd episode with a summary and discussion of last week's big AI news!
With hosts Andrey Kurenkov (https://twitter.com/andrey_kurenkov) and Jeremie Harris (https://twitter.com/jeremiecharris)
See full episode notes here.
Read out our text newsletter and comment on the podcast at https://lastweekin.ai/
If you would like to become a sponsor for the newsletter, podcast, or both, please fill out this form.
Email us your questions and feedback at contact@lastweekinai.com and/or hello@gladstone.ai
In this episode of Last Week in AI, we explore the latest advancements and debates in the AI field, including Google's release of Gemini 1.5, Meta's upcoming LLaMA 3, and Runway's Gen 3 Alpha video model. We discuss emerging AI features, legal disputes over data usage, and China's competition in AI. The conversation spans innovative research developments, cost considerations of AI architectures, and policy changes like the U.S. Supreme Court striking down Chevron deference. We also cover U.S. export controls on AI chips to China, workforce development in the semiconductor industry, and Bridgewater's new AI-driven financial fund, evaluating the broader financial and regulatory impacts of AI technologies.
 
Timestamps + links:
(00:00:00) Intro / Banter
Tools & Apps(00:03:24) Google opens up Gemini 1.5 Flash, Pro with 2M tokens to the public
(00:08:47) Meta is about to launch its biggest Llama model yet — here’s why it’s a big deal
(00:12:38) Runway’s Gen-3 Alpha AI video model now available – but there’s a catch
(00:16:28) This is Google AI, and it's coming to the Pixel 9
(00:17:30) AI Firm ElevenLabs Sets Audio Reader Pact With Judy Garland, James Dean, Burt Reynolds and Laurence Olivier Estates
(00:20:06) Perplexity’s ‘Pro Search’ AI upgrade makes it better at math and research
(00:23:12) Gemini’s data-analyzing abilities aren’t as good as Google claims

Applications & Business(00:26:38) Quora’s Chatbot Platform Poe Allows Users to Download Paywalled Articles on Demand
(00:32:04) Huawei and Wuhan Xinxin to develop high-bandwidth memory chips amid US restrictions
(00:34:57) Alibaba’s large language model tops global ranking of AI developer platform Hugging Face
(00:39:01) Here comes a Meta Ray-Bans challenger with ChatGPT-4o and a camera
(00:43:35) Apple’s Phil Schiller is reportedly joining OpenAI’s board
(00:47:26) AI Video Startup Runway Looking to Raise $450 Million

Projects & Open Source(00:48:10) Kyutai Open Sources Moshi: A Real-Time Native Multimodal Foundation AI Model that can Listen and Speak
(00:50:44) MMEvalPro: Calibrating Multimodal Benchmarks Towards Trustworthy and Efficient Evaluation
(00:53:47) Anthropic Pushes for Third-Party AI Model Evaluations
(00:57:29) Mozilla Llamafile, Builders Projects Shine at AI Engineers World's Fair

Research & Advancements(00:59:26) Researchers upend AI status quo by eliminating matrix multiplication in LLMs
(01:05:55) AI Agents That Matter
(01:12:09) WARP: On the Benefits of Weight Averaged Rewarded Policies
(01:17:20) Scaling Synthetic Data Creation with 1,000,000,000 Personas
(01:24:16) Found in the Middle: Calibrating Positional Attention Bias Improves Long Context Utilization

Policy & Safety(01:26:32) With Chevron’s demise, AI regulation seems dead in the water
(01:33:40) Nvidia to make $12bn from AI chips in China this year despite US controls
(01:37:52) Uncle Sam relies on manual processes to oversee restrictions on Huawei, other Chinese tech players
(01:40:57) U.S. government addresses critical workforce shortages for the semiconductor industry with new program
(01:42:42) Bridgewater starts $2 billion fund that uses machine learning for decision-making and will include models from OpenAI, Anthropic and Perplexity

(01:47:57) Outro

Our 172nd episode with a summary and discussion of last week's big AI news!
With hosts Andrey Kurenkov (https://twitter.com/andrey_kurenkov) and Jeremie Harris (https://twitter.com/jeremiecharris)
Read out our text newsletter and comment on the podcast at https://lastweekin.ai/
If you would like to become a sponsor for the newsletter, podcast, or both, please fill out this form.
Email us your questions and feedback at contact@lastweekinai.com and/or hello@gladstone.ai
(00:00:00) Intro / Banter
Tools & Apps
(00:03:02) Anthropic Debuts Collaboration Tools for Claude AI Assistant
(00:08:32) Google rolls out Gemini side panels for Gmail and other Workspace apps
(00:12:30) OpenAI delays rolling out its 'Voice Mode' to July
(00:15:40) OpenAI’s ChatGPT for Mac is now available to all users
(00:17:27) Waymo ditches the waitlist and opens up its robotaxis to everyone in San Francisco
(00:18:53) Figma announces big redesign with AI

Applications & Business
(00:21:37) Meet Sohu: The World’s First Transformer Specialized Chip ASIC
(00:29:42) Huawei Has Reportedly Invested Billions In An R&D Facility That Will Allow It To Develop Advanced Chipmaking Machinery Similar To ASML & Others
(00:32:17) China's ByteDance working with Broadcom to develop advanced AI chip, sources say
(00:35:35) Chinese AI firms woo OpenAI users as US company plans API restrictions
(00:39:45) OpenAI walks back controversial stock sale policies, will treat current and former employees the same

Projects & Open Source
(00:43:42) Meta Large Language Model Compiler: Foundation Models of Compiler Optimization
(00:47:54) Google’s Gemma 2 series launches with not one, but two lightweight model options—a 9B and 27B
(00:48:50) ESM3: Simulating 500 million years of evolution with a language model

Research & Advancements
(00:56:57) Finding GPT-4’s mistakes with GPT-4
(01:03:30) Chinese-built ChatGLM exceeds GPT-4 Across Several Benchmarks
(01:07:15) Performances are plateauing, let's make the leaderboard steep again
(01:11:18) Structural mechanism of bridge RNA-guided recombination
(01:15:01) Reconciling Kaplan and Chinchilla Scaling Laws

Policy & Safety
(01:17:42) Safety Alignment Should Be Made More Than Just a Few Tokens Deep
(01:23:02) Y Combinator rallies start-ups against California’s AI safety bill
(01:28:20) Pro-Kigali propagandists caught using Artificial Intelligence tools
(01:21:40) Coordinated Disclosure of Dual-Use Capabilities: An Early Warning System for Advanced AI
(01:35:08) Adversaries Can Misuse Combinations of Safe Models
Mitigating Skeleton Key, a new type of generative AI jailbreak technique

Synthetic Media & Art
(01:39:35) Music labels sue AI music generators for copyright infringement
(01:42:43) YouTube is trying to make AI music deals with major record labels
(01:45:07) Toys ‘R’ Us Debuts First Video Ad Using Sora, OpenAI’s Text-to-Video Tool

(01:49:12) Outro + AI Song

Our 171st episode with a summary and discussion of last week's big AI news!
With hosts Andrey Kurenkov (https://twitter.com/andrey_kurenkov) and Jeremie Harris (https://twitter.com/jeremiecharris)
Feel free to leave us feedback here.
Read out our text newsletter and comment on the podcast at https://lastweekin.ai/
Email us your questions and feedback at contact@lastweekin.ai and/or hello@gladstone.ai
Timestamps + Links:
(00:00:00) Intro / Banter
Tools & Apps(00:03:13) Apple Intelligence: every new AI feature coming to the iPhone and Mac
(00:10:03) ‘We don’t need Sora anymore’: Luma’s new AI video generator Dream Machine slammed with traffic after debut
(00:14:48) Runway unveils new hyper realistic AI video model Gen-3 Alpha, capable of 10-second-long clips
(00:18:21) Leonardo AI image generator adds new video mode — here’s how it works
(00:22:31) Anthropic just dropped Claude 3.5 Sonnet with better vision and a sense of humor

Applications & Business(00:28:23 ) Sam Altman might reportedly turn OpenAI into a regular for-profit company
(00:31:19) Ilya Sutskever, Daniel Gross, Daniel Levy launch Safe Superintelligence Inc.
(00:38:53) OpenAI welcomes Sarah Friar (CFO) and Kevin Weil (CPO)
(00:41:44) Report: OpenAI Doubled Annualized Revenue in 6 Months
(00:44:30) AI startup Adept is in deal talks with Microsoft
(00:48:55) Mistral closes €600m at €5.8bn valuation with new lead investor
(00:53:12) Huawei Claims Ascend 910B AI Chip Manages To Surpass NVIDIA’s A100, A Crucial Alternative For China
(00:56:58) Astrocade raises $12M for AI-based social gaming platform

Projects & Open Source(01:01:03) Announcing the Open Release of Stable Diffusion 3 Medium, Our Most Sophisticated Image Generation Model to Date
(01:05:53) Meta releases flurry of new AI models for audio, text and watermarking
(01:09:39) ElevenLabs unveils open-source creator tool for adding sound effects to videos

Research & Advancements(01:12:02) Samba: Simple Hybrid State Space Models for Efficient Unlimited Context Language Modeling
(01:22:07) Improve Mathematical Reasoning in Language Models by Automated Process Supervision
(01:28:01) Introducing Lamini Memory Tuning: 95% LLM Accuracy, 10x Fewer Hallucinations
(01:30:32) An Empirical Study of Mamba-based Language Models
(01:31:57) BERTs are Generative In-Context Learners
(01:33:33) SELFGOAL: Your Language Agents Already Know How to Achieve High-level Goals

Policy & Safety(01:35:16) Sycophancy to subterfuge: Investigating reward tampering in language models
(01:42:26) Waymo issues software and mapping recall after robotaxi crashes into a telephone pole
(01:45:53) Meta pauses AI models launch in Europe
(01:46:44) Refusal in Language Models Is Mediated by a Single Direction
Sycophancy to subterfuge: Investigating reward tampering in language models
(01:51:38) Huawei exec concerned over China’s inability to obtain 3.5nm chips, bemoans lack of advanced chipmaking tools

Synthetic Media & Art(01:55:07) It Looked Like a Reliable News Site. It Was an A.I. Chop Shop.
(01:57:39) Adobe overhauls terms of service to say it won’t train AI on customers’ work
(01:59:31) Buzzy AI Search Engine Perplexity Is Directly Ripping Off Content From News Outlets

(02:02:23) Outro + AI Song 

Our 170th episode with a summary and discussion of last week's big AI news!
With hosts Andrey Kurenkov (https://twitter.com/andrey_kurenkov) and Jeremie Harris (https://twitter.com/jeremiecharris)
Feel free to leave us feedback here.
Read out our text newsletter and comment on the podcast at https://lastweekin.ai/
Email us your questions and feedback at contact@lastweekin.ai and/or hello@gladstone.ai
Timestamps + Links:
Tools & Apps(00:03:33) KLING is the latest AI video generator that could rival OpenAI's Sora
(00:09:16) ‘Apple Intelligence’ will automatically choose between on-device and cloud-powered AI
(00:12:21) Udio introduces new udio-130 music generation model and more advanced features
(00:14:38) Perplexity AI’s new feature will turn your searches into shareable pages
(00:16:35) ElevenLabs’ AI generator makes explosions or other sound effects with just a prompt
(00:18:37) Google’s updated AI-powered NotebookLM expands to India, UK and over 200 other countries

Applications & Business(00:19:40) OpenAI is restarting its robotics research group
(00:25:01) Saudi fund invests in China effort to create rival to OpenAI
(00:29:34) UAE seeks ‘marriage’ with US over artificial intelligence deals
(00:33:01) Zoox to test self-driving cars in Austin and Miami 
(00:35:49) Microsoft Lays Off 1,500 Workers, Blames "AI Wave"
(00:38:28) Avengers, assemble—Google, Intel, Microsoft, AMD and more team up to develop an interconnect standard to rival Nvidia's NVLink

Projects & Open Source(00:40:39) GLM-4-9B-Chat-1M
(00:46:37) Hugging Face and Pollen Robotics show off first project: an open source robot that does chores
(00:49:40) Zyphra debuts Zyda, a 1.3T language modeling dataset it claims outperforms Pile, C4, arxiv
(00:51:59) Stability AI debuts new Stable Audio Open for sound design

Research & Advancements(00:54:05) Scaling and evaluating sparse autoencoders
(01:04:54) Improving Alignment and Robustness with Short Circuiting
(01:12:11) Automatic Data Curation for Self-Supervised Learning: A Clustering-Based Approach
(01:16:20) GPT-4 didn't ace the bar exam after all, MIT research suggests — it didn't even break the 70th percentile

Policy & Safety(01:20:11) Former OpenAI researcher foresees AGI reality in 2027
(01:28:03) OpenAI Insiders Warn of a ‘Reckless’ Race for Dominance
(01:33:52) Testing and mitigating elections-related risks
(01:36:26) Teams of LLM Agents can Exploit Zero-Day Vulnerabilities

Synthetic Media & Art(01:43:23) The Uncanny Rise of the World's First AI Beauty Pageant

(01:46:25) Outro + AI Song