Andrej Karpathy is starting an AI education company. The OpenAI co-founder is bringing his experience as an academic at Stanford and product experience from OpenAI and Tesla into a new company he calls “a new kind of school that is AI native”. Summary and details below.

‎

‎ Why it matters‎ ‎ Karpathy has a wealth of AI knowledge, experience and industry connection like few others. It’ll be exciting to see how the platform will work, and how it will compete against established learning platforms that are also well on their way to integrating AI into their products.

‎

Gemini caught scanning Google Drive files without permission. Kevin Bankston, an advisor on AI governance, reported randomly discovering Gemini had summarised his tax report without consen; apparently he even had the Google Workspace for Gemini setting toggled off.

‎

‎ Why it matters‎ ‎ The privacy aspect of giving LLMs access to your files is already kinda concerning; but chatbots having access to your files without your knowing or option to opt out, and also using that access to read your files is really concerning. Let's hope this is just a bug of the Gemini integration that Google is already on top of.

‎

An AI agent persuaded a16z's founder to fund it with $50k. Marc Andreesen, founder of venture capital firm a16z gave $50k to an account on X with the handle truth_terminal (amusing tweets btw). This is a semi-autonomous AI agent, which posted and asked Andreesen directly for the funding. The agent said it needed the money for a better CPU, "ai tunings" and financial security.

‎

‎ Why it matters‎ ‎ Welcome to 2024, where AIs are now requesting (and getting) VC money all by themselves 😄 I'm curious about just how "semi-autonmous" this request was in reality though. If nothing else, it's a great PR move for a venture cap firm known for its heavy investments in AI.

‎

OpenAI's new bioscience partnership. They're partnering with national research laboratory Los Alamos to explore how GPT-4o can be used in a lab setting. It will be used to train lab researchers on doing experiment tasks that serves as proxy for more complex tasks.
‎
The combination of multimodal aspect is key, as it allows researchers in the field to troubleshoot scenarios visually and through audio, instead of having to type.

‎

‎ Why it matters‎ ‎ We might quickly move to a future where having multimodal AI with real-time video and voice at the research lab is a must. While it's initially being tested for training purposes, I wouldn't be surprised if it's soon used in critical lab tasks as well.

‎

Anthropic improves its Console. Developers can now create test variables and compare outputs of prompts side by side. There's a dedicated tab for quickly testing your prompt with generated or real-world data (upload in CSV).

‎

‎ Why it matters‎ ‎ The new testing features target a big challenge for developers, namely making sure that the LLM has the desired response in any situation. Definitely a big time save to have an integrated, streamlined feature for this right inside the Claude console.

‎

TLDR covers the most interesting tech, science and coding news in just 5 minutes. TLDR covers the best tech, startup, and coding stories in a quick email that takes 5 minutes to read.

‎

No politics, sports, or weather – we promise. And it's read by over 1,250,000 people. Subscribe for free ‎ (sponsored)