Skip to main content
Back

Thinking Machines TML-Small 64.7%, MIT Brain Study , Rust Browser

Thinking Machines ships real-time AI, open-source video model runs locally, Rust browser loads in 85ms, Meta halves LLM decoding

May 13, 2026 6 min read

Hey,

AI is getting better at talking to us. But MIT just reminded us we might be getting worse at thinking.

Thinking Machines ships a model that listens, speaks, and reasons simultaneously, closing the gap between human conversation and AI interaction. Meanwhile, science confirms the more we outsource thinking to AI, the less our brains actually fire.

The irony is almost too clean. We're building smarter interfaces while quietly accumulating what researchers call "cognitive debt." Worth sitting with.
Together with · partner_image
Author photo
Today's Author
Lior Alexander. Founder of AlphaSignal and former ML Engineer. Previously built ML systems at Iguazio, Guesty, Enphase, Mila.
Summary

Read time: 6 min 46 sec

Top News

▸ Thinking Machines releases a real-time AI model that talks, listens, and thinks simultaneously

Weights & Biases

▸ New guide: Tools & workflows to develop AI agents

Top Paper

▸ Study finds ChatGPT users show weaker brain activity and own their writing less

Wispr Flow

▸ One voice input. Every device. Every app.

Top Model

▸ Uncensored open-source video model runs locally, generates 10-second clips at 24fps

Signals

1. Open-source Rust browser loads pages in 85ms using 10x less RAM than Chrome
2. Presented by Viktor: Run a 10-person team that operates like 200. Viktor lives in Slack, runs 3,000 tools, does real work.
3. Open-source proxy lets you use free Claude models from Amazon's Kiro in any AI coding tool
4. Meta FAIR's new byte-level model cuts LLM decoding steps in half
5. Kaiming He's new diffusion model generates text in continuous space, not discrete tokens
6. One developer used an AI agent to rank the top 10 LLM learning resources
Top News
Thinking Machines releases a real-time AI model that talks, listens, and thinks simultaneously
20,905 Likes
Thinking Machines releases a real-time AI model th

Most AI assistants work like walkie-talkies. You talk, they wait, then they respond. Thinking Machines just flipped that model entirely.

Their new TML-Interaction-Small is a 276-billion parameter model that listens, speaks, and watches video all at the same time, in real time. No waiting for your turn to finish. It processes input and output in 200ms chunks simultaneously, the way humans actually converse.

Under the hood, there are two parts working together:

  • An interaction layer that stays live in the conversation, handling timing, cues, and natural back-and-forth
  • A background layer that quietly handles web searches, reasoning, and tool calls, then feeds results into the conversation without breaking the flow

The benchmarks are striking. On timed speech tasks, it scores 64.7% vs. 4.3% for GPT-Realtime-2. It also tracks when you are thinking, pausing, or inviting a response, with no hardcoded dialogue rules. A technical report with full evaluation results is available now.

Read more
forward →
Presented by Weights & Biases
New guide: Tools & workflows to develop AI agents

AI agents can dramatically boost productivity and innovation, but getting them into the real world takes a lot of iteration.

Whether you’re exploring agents for the first time or refining your current approach, this primer delivers actionable insights to help your team succeed and thrive in the AI era.

Get the guide to learn:

  • What defines agentic applications and why observability matters
  • How pioneering companies are building and deploying AI agents
  • A proven workflow for building agentic AI applications
Get the guide
partner with us →
Top Paper
Study finds ChatGPT users show weaker brain activity and own their writing less
15,495 Likes
Study finds ChatGPT users show weaker brain activi

An MIT study put AI writing tools under a brain scanner, literally. Researchers strapped EEG headsets (devices that read electrical brain activity) on 54 participants across three groups: ChatGPT users, search engine users, and no-tool writers. They measured what was actually happening inside people's heads over four months.

The results are uncomfortable. The brain-only group showed the strongest neural networks, while the LLM group exhibited up to 55% reduced connectivity. Less brain activity means less thinking and less memory formation.

The wildest part: 83% of LLM users couldn't quote from the essays they had just written.

Key findings:

  • Researchers coined "cognitive debt" to describe how AI spares mental effort short-term but causes long-term costs including diminished critical thinking and shallow information processing.
  • When brain-only users switched to AI, they showed increased brain connectivity, suggesting AI used properly could enhance learning.
  • Researchers recommend delaying AI integration until learners have first engaged in sufficient self-driven cognitive effort.

The takeaway: use AI as a finishing tool, not a starting one.

Read more
forward →
Presented by Wispr Flow
One voice input. Every device. Every app.

Speak a prompt into Cursor on your laptop. Reply to a reviewer on your phone. Dictate a methodology section on Windows at the office.

Wispr Flow works the same in every app on Mac, Windows, iOS, and Android.

89% sent with zero edits. Used by teams at OpenAI and Vercel.

Try flow
partner with us →
Top Model
Uncensored open-source video model runs locally, generates 10-second clips at 24fps
4,388 Likes
Uncensored open-source video model runs locally, g

A new open-source video AI just dropped, and it does something commercial tools refuse to do.

Sulphur 2 is built on Lightricks' LTX video model, but fine-tuned with no content filters. That means you can generate videos that platforms like Runway or Sora would instantly reject.

Here's what it can do:

  • Turn text or images into 10-second, 24fps realistic videos
  • Run fully local on your machine, no subscription, no usage logs
  • Handle violent, adult, or extreme content with no restrictions (illegal content excluded)
  • Trained on 125,000+ real videos totaling over 500GB

To run it, download the model from HuggingFace and load it into ComfyUI or Pinokio. You need at least 16GB VRAM, an RTX 4090 runs it comfortably.

What this opens up: niche stock video markets, uncensored creative projects, and fine-tuning your own specialized model on top of it.

Read more
forward →
Signals
1
Open-source Rust browser loads pages in 85ms using 10x less RAM than Chrome
1,976 Likes
2
Run a 10-person team that operates like 200. Viktor lives in Slack, runs 3,000 tools, does real work.
Presented by Viktor
3
Open-source proxy lets you use free Claude models from Amazon's Kiro in any AI coding tool
1,314 Stars
4
Meta FAIR's new byte-level model cuts LLM decoding steps in half
704 Likes
5
Kaiming He's new diffusion model generates text in continuous space, not discrete tokens
603 Likes
6
One developer used an AI agent to rank the top 10 LLM learning resources
1,945 Likes
At Alpha Signal, our mission is to build a sharp, engaged community focused on AI, machine learning, and cutting-edge language models, helping over 200,000 developers stay informed and ahead. We're passionate about curating the best in AI, from top research and trending technical blogs to expert insights and tailored job opportunities. We keep you connected to the breakthroughs and discussions that matter, so you can stay in the loop without endless searching. We also work closely with partners who value the future of AI, including employers and advertisers who want to reach an audience as passionate about AI as we are.

Our partnerships are based on shared values of ethics, responsibility, and a commitment to building a better world through technology.Privacy is a priority at Alpha Signal. Our Privacy Policy clearly explains how we collect, store, and use your personal and non-personal information. By using our website, you accept these terms, which you can review on our website. This policy applies across all Alpha Signal pages, outlining your rights and how to contact us if you want to adjust the use of your information. We're based in the United States. By using our site, you agree to be governed by U.S. laws.
Looking to promote your company, product, service, or event to 250,000+ AI developers?
Work with us