Skip to main content
Back

Meta’s AI Reads Your Mind Watching Movies

Vercel upgrades to v0.app, Claude gets memory, OpenAI tops IOI, Gemini adds Deep Think, Perplexity adds video.

August 12, 2025 4 min read

Hey

Welcome to AlphaSignal, the most read source of news by AI engineers and researchers. Every day, we identity and summarize the top 1% of news, papers, models, and repos, so you’re always up to date.


Here’s today’s roundup:

./SUMMARY

Read time: 4 min 37 sec

Top News

▪ Meta builds TRIBE to predict brain activity from movies

Galileo

▪ Build automated AI evaluation pipelines with Galileo’s free 70-page guide

Trending Signals

▪ Vercel launches agentic AI upgrade for v0.app app builder

▪ Anthropic adds past chat memory to Claude for higher-tier plans

▪ Google ships Deep Think to Gemini app for Ultra subscribers

▪ OpenAI's reasoning model ranks #1 among AI at International Olympiad in Informatics

▪ Perplexity rolls out video generation across all platforms

Tenex

▪ Apply now to lead AI strategy and drive enterprise revenue growth

Trending Repos

▪ dyad: Local, privacy-focused AI app builder

▪ ai-engineering-hub: Tutorials and projects on LLMs, RAG, AI agents

▪ umami: Self-hosted web analytics with real-time dashboards

How To

▪ Build and preview frontends with GPT-5 by OpenAI

If you're enjoying AlphaSignal please forward this email to a colleague. 

It helps us keep this content free.

./TOP_NEWS

Meta publishes TRIBE, combining video, audio, and text to outperform single-modality brain modeling approaches by 30%

2,478 Likes

Meta’s FAIR team has built TRIBE (Trimodal Brain Encoder), a 1-billion-parameter neural network that predicts how the human brain responds while watching movies. It processes video, audio, and text at the same time.


It took first place in the Algonauts 2025 brain-modeling competition with a large margin over visual-only or audio-only models..


Key Features

  • First trimodal model for brain response prediction.

  • Predicts spatial and temporal fMRI activation patterns without direct brain scanning.

  • Works across ~1,000 brain regions, including areas integrating sight, sound, and language.

  • Correctly predicted over 50% of brain activity patterns across the brain regions.

  • Outperforms single-modality models by ~30% in high-level associative regions.

  • Demonstrated consistent accuracy across multiple subjects, not just group-averaged data.

How It Works

  • Extracts features from video, audio, and dialogue using frozen pretrained models for each modality.

  • Aligns and concatenates features across time frames.

  • Processes the sequence with a transformer brain encoder that learns to predict voxel-level fMRI signals.

  • Trains only the brain encoder on ~80 hours of aligned fMRI and movie data.

Access and Availability

  • Source code available in github repo.

  • Download the data from: https://cneuromod.ca

  • License: CC BY-NC 4.0 for research and non-commercial use.

  • Pretrained models and training scripts included.

  • Requires compatible GPU hardware for training and inference.

Github repo

Unlock 10x Faster AI Evaluation with Galileo’s Free 70-Page Guide

Stop relying on vibe checks. Galileo’s new eBook shows you how to build automated AI evaluation pipelines using LLM judges. Get code, frameworks, and real-world methods in 70 pages of actionable content.

  • Automate scoring, explanations, and quality issue flagging

  • Use token-level scoring and Chain-of-Thought for precision

  • Apply pairwise comparisons to validate outputs at scale

  • Address LLM bias: verbosity, authority, position

  • Implement with provided frameworks and clear code samples

Learn how experts quickly and accurately run evals.

Get the free ebook

partner with us

./TOP_SIGNALS

Vercel rebrands v0.dev to v0.app, adding agentic AI that plans, builds, and deploys complete apps

1,243 Likes

Anthropic’s Claude gains the ability to recall past chats for Max, Team, and Enterprise users

6,948 Likes

Google rolls out Deep Think to Gemini Ultra users, a high-performance reasoning model for math, coding, and logic tasks

2,649 Likes

OpenAI’s general reasoning system ranks 6th overall and 1st among AI at IOI without competition-specific fine-tuning

2,485 Likes

Perplexity launches video creation on web and mobile, offering 5–15 videos per month by tier

962 Likes

Lead high-ROI AI initiatives and scale revenue as Tenex’s Technical Strategist

Work on-site in New York City to run AI strategy audits, design product roadmaps, automate workflows, and manage C-suite relationships. Earn $160K–$500K annually, directly tied to the revenue you generate.


Join Tenex and start driving measurable AI impact today.

Apply to lead AI strategy at Tenex ↗️

./TRENDING_REPOS

dyad

12,438 Stars

Dyad is an open-source AI app builder that runs locally on macOS and Windows. It focuses on speed and privacy, works offline, and lets you integrate your own API keys for custom model use without cloud dependency.

ai‑engineering‑hub

16,839 Stars

This repo provides tutorials, guides, and projects on LLMs, retrieval-augmented generation, and AI agents. It includes practical examples, code, and workflows to help implement and understand AI systems for real-world applications.

umami

29,548 Stars

It is a self-hosted web analytics tool that tracks website usage without cookies. It offers real-time dashboards, trend comparisons, CSV exports, and multi-site support, focusing on privacy and speed under the MIT license.

./HOW_TO

Build and preview frontends with GPT-5

What you build

You generate HTML (and JS/CSS when requested) with GPT-5, extract the code block from the response, save it to a file, and open it in your browser. You can also pass an image of an existing UI to match its style.


Prerequisites

  • Python and openai SDK installed.

  • OPENAI_API_KEY set in your environment.

  • GPT-5 API access

Step 1

Set your API key.


export OPENAI_API_KEY=sk-...

Step 2

Create a file frontend_gpt5.py and paste the helper functions from the guide below.


Step 3

Generate a landing page:


from frontend_gpt5 import make_site
make_site("Create a landing page for a retro games
              store in a noir theme.", "retro_dark.html")

Step 4

Open the generated HTML file in your browser.


Step 5

Modify the theme or add interactivity by changing the prompt. You can also pass an image to match an existing UI style.

Openai's guide

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 200,000+ AI developers? Let's work together.

Work with us