🥟 Chao-Down #316 The rise of "anti-woke" AI, Boston Dynamics retires Atlas humanoid robot, Drake drops song with an AI Tupac Shakur verse, SF exec explains layoff trend as a way to free cash for AI

An interesting piece came in from the Wall Street Journal titled “Here Come the Anti-Woke AIs” which explains how the release of Metas’ Llama-3 and more broadly open-source models improve accessibility but lack the safety and responsible AI guardrails that closed-source systems are required to have.

Some people see this as good as we’ve seen very public failures (i.e. Google’s Gemini) of bias creeping into closed AI models.

But what’s more interesting to me is the reality that many of these AI models reflect the values of those who train them. By releasing the models in the open-source, people (who know how) can fine-tune highly personal AIs that actually can do and say what they want as opposed to a filtered version offered by a big tech company.

So yes, we may see more “anti-woke” AIs running around. But even more so, I anticipate a rise in AIs that can more accurately reflect people’s individual preferences and values.

The question is: will that all be a net positive?

-Alex, your resident Chaos Coordinator.

What happened in AI? 📰

Boston Dynamics Retires Its Legendary Humanoid Robot (IEEE Spectrum)

GPT-4 can exploit real vulnerabilities by reading advisories (The Register)

Microsoft’s AI Copilot Is Starting to Automate the Coding Industry (Bloomberg)

SF exec defends 'brutal' trend: Lay off staff to free cash for AI (sfgate.com)

Drake Drops 'Taylor Made Freestyle' With AI Tupac Shakur Verse (Variety)

People are afraid of self-driving cars — can the industry change that? (The Verge)

Always be Learnin’ 📕 📖

Embeddings are a good starting point for the AI curious app developer (Substack)

Being busy and the illusion of productivity (Ness Labs)

Live or Die by Product-Market Fit (OnlyCFO's Newsletter)

Projects to Keep an Eye On 🛠

meta-llama/PurpleLlama: Set of tools to assess and improve LLM security. (Github)

phidatahq/phidata: Build AI Assistants with memory, knowledge and tools. (Github)

facebookresearch/llm-transparency-tool: LLM Transparency Tool (LLM-TT), an open-source interactive toolkit for analyzing internal workings of Transformer-based language models. (Github)

The Latest in AI Research 💡

Compression Represents Intelligence Linearly (arxiv)

Reka Core, Flash, and Edge: A Series of Powerful Multimodal Language Models (arxiv)

Toward Self-Improvement of LLMs via Imagination, Searching, and Criticizing (arxiv)

The World Outside of AI 🌎

U.S.-China Internet War Intensifies as House Passes TikTok Ban (WSJ)

Google's Top Secret Camera Lab Is Like an Ikea for Pixel Testing (CNET)

Sedentary lifestyles are a ticking health timebomb (The Guardian)

Google tells staff not to 'debate politics' after firings in Israel protest (qz.com)

Private equity finds its next bet: college admissions (CNN Business)

Samsung shifts to six-day workweeks to ‘inject a sense of crisis’ (The Verge)

One Last Bite 😋