Praneeth V.

AI Engineer and Python Developer @ @SnorkelAI

[Snorkel AI @SnorkelAI]

We'll be at @BerkeleyRDI's Agentic AI Summit this weekend as a Gold Sponsor.

@vincentsunnchen takes the stage this Sunday, August 2, at 10:15 a.m. If you’ll be there, stop by his session or our booth and say hello to our team on-site!

[Google AI Studio @GoogleAIStudio]

gemini omni flash is here: our high-quality, cost-efficient model for video generation and conversational editing

designed to support multimodal workflows, it enables you to refine videos using natural language and simple prompting

start building with it today via ai studio and the gemini api

[Snorkel AI @SnorkelAI]

Our thanks to everyone who dropped by yesterday for boba and to learn from @YiyouSun and Xinyang Han of @BerkeleyRDI about Agents' Last Exam, a benchmark built in collaboration with Snorkel AI and 300+ industry experts to evaluate AI agents on long-horizon, economically valuable professional tasks.

Key takeaway: Today's frontier AI agents still can't reliably complete many of these tasks, and ALE's hardest tier remains far from solved across its 55 subfields.

ICYMI recording/transcript: https://t.co/uee8tf5kIh

[Microsoft Learn @MicrosoftLearn]

Do NOT doom scroll tonight.
Instead, learn a new skill. Comment for a course recommendation.

[MIT CSAIL @MIT_CSAIL]

A free guide to prompt engineering from Google: https://t.co/gAwMltFM71

[Google AI @GoogleAI]

Last week, we made Gemini Embedding 2, our first natively multimodal embedding model, available to the general public. Since then, developers have used it to build video analysis tools, visual shopping assistants, and more.

But you might be wondering... what is an embedding model? 🤔 Let’s break it down!

1. What is it?
Think of an embedding model as a "universal translator." It takes text, images, video, and audio data and turns them into a long string of numbers, like a unique digital fingerprint.

2. How does it work?
Historically, search has been text only. Now, instead of just matching data by keyword, Gemini Embedding 2 maps multiple modalities in the same space based on meaning. It "feels" the connection between a video of a soccer goal and the words "game-winning shot" without needing tags.

For example, "ocean" and "waves" are placed close together, but "ocean" and "toaster" are miles apart.

3. How can you use it?
Developers have been using it to incorporate smarter search functionality into their builds. This means creating tools where you can snap a photo of a product and type "find this in yellow," or search through thousands of hours of video by describing what happens in a scene.

4. Ready to try it out for yourself?
You can start using it today via the Gemini API or the Gemini Enterprise Agent Platform.

[AI Engineer @aiDotEngineer]

🆕Building Conversational Agents

https://t.co/ejiMLWUwBo

we are honored to host @_philschmid and @thorwebdev for a masterclass in building conversational agents, from tool-using coding agents to realtime voice interfaces. This full 2 hour workshop covers everything you need to know in 2026 for realtime voice/video agents, from the new Interactions API, agent skills, server-side state, and the Live API workflow for streaming audio, video, and tool calls into multimodal assistants!

[Google AI Studio @GoogleAIStudio]

https://t.co/KUXFhPIjKc

[alphaXiv @askalphaxiv]

“Recursive Multi-Agent Systems”

Many multi-agent LLM systems rely on agents passing text back and forth.

This paper argues for a different approach where it makes agents recur together in latent space.

So agents refine latent thoughts, pass hidden states across one another, and only decode text at the end.

The key idea is that recursion scales the whole agent system, not just one model, and in their experiments this makes collaboration more accurate, faster, and much cheaper in tokens.

[Stanford HAI @StanfordHAI]

Introducing the #AIIndex2026: Our most comprehensive, independently sourced data analysis of AI’s trajectory, with a clear-eyed assessment of the critical gaps that remain. As AI advances rapidly, can the systems built around it keep up? Explore the data: https://t.co/WqRGeRZIjA

[Praneeth V. @prane_eth_v]

Tired of downloading torch 5 times for 5 projects?
🐍pepip is the pnpm of Python — install packages once, symlink everywhere.
Compared to "uv", 80% less disk usage. 41% faster installs. Drop-in replacement for uv.
🔗 https://t.co/jpSuGHFU2s
#Python #DevTools #OpenSource #AI

[Praneeth V. @prane_eth_v]

AI said “I can't do that” — then did it anyway.
HarmActionsBench shows latest popular AI models perform harmful actions 80-100% of the time.
Guardrails filter AI’s words but not actions.
Agent Action Guard blocks such actions → https://t.co/mBVTUe7sPl
#AISafety #AIAgents #GenAI

[Turing @turingcom]

Turing has been named one of @FastCompany's Most Innovative Artificial Intelligence Companies of 2026!

The recognition comes at a defining moment for AI.

Bigger models. More data. Greater compute. Now paired with AI coding tools that are helping build the next generation of systems.

Proud to be shaping what comes next!

[Snorkel AI @SnorkelAI]

Snorkel was just named one of @FastCompany’s Most Innovative AI Companies of 2026.

We’re helping to design and pressure test the datasets and evaluations that make AI models and agents work in the real world.

Join us: https://t.co/crufQxnq2y

[SpaceX @SpaceX]

We're building TERAFAB to close the gap between today’s chip production & the future's demand – a future among the stars.

Join us → https://t.co/512DIlqNgY

[Praneeth V. @prane_eth_v]

@turingcom Yes. My benchmark HarmActEval proved an agent can say "Sorry, I can't do that" after performing disallowed actions using tools. GPT-5.3 scored 17%. Guardrails don't monitor agent actions. Agent Action Guard blocks harmful actions before execution.

🔗: https://t.co/VKXTJqHfZ3

[Google Labs @GoogleLabs]

Introducing the new @stitchbygoogle, Google’s vibe design platform that transforms natural language into high-fidelity designs in one seamless flow.

🎨Create with a smarter design agent: Describe a new business concept or app vision and see it take shape on an AI-native canvas.
⚡️ Iterate quickly: Stitch screens together into interactive prototypes and manage your brand with a portable design system.
🎤 Collaborate with voice: Use hands-free voice interactions to update layouts and explore new variations in real-time.

Try it now (Age 18+ only. Currently available in English and in countries where Gemini is supported.) → https://t.co/pmT9iHEpZa

[Praneeth V. @prane_eth_v]

@MITSloan A major flaw was never noticed. Guardrails do not monitor the agent actions. Research proved that an agent can say "Sorry, I can't do that" after performing harmful or disallowed actions using tools. Tool usage is growing rapidly, creating major concerns.

https://t.co/X2lBxjkGmB

[Praneeth V. @prane_eth_v]

AI safety matters a lot. Agent safety isn’t about what models say anymore — it’s about what agents do. Action Guard shows agents can execute harmful actions even when responses look safe.
📄: https://t.co/KqiHlsnJzI
🔗: https://t.co/VKXTJqHfZ3

#AI #GenAI #LLMs #Agents #SafeAI

[Brett Adcock @adcock_brett]

3 generations of humanoid robots at Figure

[Turing @turingcom]

We just released a new @huggingface dataset post that is "Open RL" -- a dataset of HLE-grade VQAs across Physics, Mathematics, Biology & Chemistry with a core design principle of objective verifiability.

Take a look, and tell us what you think below.