SunBrief#74:Meta reveals AI that predicts brain activity

Welcome to the SunBrief

Today in SunBrief 🌞

• 100+ ChatGPT Prompts to Revolutionize Your Day

• Meta’s TRIBE v2 Predicts Brain Activity From Video, Audio, and Text

• Apple Is Opening Siri To Rival AI Assistants

• Stock Updates

• Iran-Linked Hackers Claim Breach of FBI Director Kash Patel’s Personal Email

• AI Highlights of the Week

• Too Important to Miss

100+ ChatGPT Prompts to Revolutionize Your Day

Stop wasting time figuring out what to ask ChatGPT. This free HubSpot guide gives you 100+ proven prompts to speed up writing, research, planning, and daily tasks.

From better outputs to faster workflows, this is a simple shortcut to getting more value from ChatGPT. Instant access. Real use cases. Prompts you’ll actually use.

Meta’s TRIBE v2 Predicts Brain Activity From Video, Audio, and Text

FAIR’s new model uses large-scale fMRI training to simulate how the brain responds to real-world stimuli

Meta’s AI lab released TRIBE v2, a model trained on 1,000+ hours of fMRI data from 720 people that predicts how the brain responds to images, sounds, and language, often matching typical brain responses better than a single scan.

Key Points:

• Tri-modal brain prediction: TRIBE v2 takes video, audio, and text, converts each into embeddings via Video-JEPA-2 (video), Wav2Vec-BERT-2.0 (audio), and Llama 3.2 (text), then predicts full-brain activity across ~70,000 voxels.

• Less noise than single scans: Because fMRI readings are noisy, the model often correlates more strongly with the group-average response than individual scans do, especially on higher-quality datasets.

• Replays classic neuroscience results: In controlled tests, it reproduced known findings like specialized regions for faces, places, bodies, and language, suggesting experiments could be tested in silico before lab work.

• Shows which senses drive which regions: Turning channels on/off revealed expected mappings (audio→auditory cortex, video→visual cortex, text→language/frontal areas), and multi-sensory regions benefited most from combining inputs.

• Clear limits remain: It still treats the brain as a passive receiver, only covers three modalities, and inherits fMRI’s slow timing (blood-flow delays), so it can’t capture fast neural dynamics or action/decision loops.

Why It Matters:
If models like TRIBE v2 keep improving, they could reduce the cost and time of brain research, enable testing hypotheses before human scans, and accelerate brain-inspired AI and future clinical use, while still needing better coverage of behavior, more senses, and higher time resolution signals.

Apple Is Opening Siri To Rival AI Assistants

Apple expands Siri to support ChatGPT, Gemini, Claude, and more through a new Extensions system

Apple may turn Siri into a broader platform in iOS 27, allowing third-party AI chatbots to plug into Siri and letting users choose which services handle different requests. That would make Siri more of a distribution layer than Apple’s defining AI model moment.

Key Points:

• Siri becomes a switchboard: A new Extensions system would let installed AI apps work inside Siri across iOS, iPadOS, and macOS, instead of relying on one exclusive partner.

• End of exclusivity: ChatGPT would lose its exclusive Siri position (from 2024) as Gemini, Claude, and others gain the same access.

• Marketplace mechanics: Apple would reportedly add an App Store section for AI services and let any chatbot integrate through an API, widening the funnel.

• Services revenue angle: The model implies Apple can keep earning App Store commissions on AI subscriptions, even if its own Siri intelligence lags rivals.

• Parallel “new Siri” still planned: Apple is still said to be working on its own chatbot-style Siri overhaul (conversational threads, file analysis, screen awareness), but delays and past misses make Extensions a hedge.

Why It Matters:

This reframes Apple’s AI strategy around controlling the entry point. Siri does not need to beat every model. It just needs to remain the default doorway where models compete, while Apple owns the interface and the revenue flow.

Stock Updates

Iran-Linked Hackers Claim Breach of FBI Director Kash Patel’s Personal Email

Handala posts alleged Gmail cache; FBI says the material is “historical” and contains no government info

A pro-Iran hacking group known as Handala says it accessed FBI Director Kash Patel’s personal email and published a cache of files it claims came from that account. The FBI acknowledged the targeting and said it has taken steps to mitigate risk, emphasizing the exposed information is historical and not government-related.

Key Points:

• Claimed leak: Handala posted what appears to be years of emails, photos, and personal documents tied to Patel’s alleged Gmail address.

• Timeframe: Reported materials largely date from roughly 2010–2019.

• Official response: The FBI said it’s aware of the activity, is mitigating risks, and that the information involves no government information.

• Broader context: Handala has recently claimed other disruptive hacks, including against major companies, amid escalating geopolitical tensions.

Why It Matters:
This incident highlights how personal accounts can become high-impact targets and how leaked “non-government” data can still create security, reputational, and operational risk through exposure of contacts, travel, and personal history.

AI Highlights of the Week

• Wikipedia Bans AI-Generated Article Writing
  Wikipedia introduces a new policy, prohibiting AI-generated content in articles while still allowing limited use in editing workflows.
  Editors can use AI for basic copyedits with human review, but not for creating or rewriting content, ensuring accuracy, reliability, and source-backed information remain intact.

• Anthropic Wins Court Battle Against US Government
  A federal judge rules in favor of Anthropic, blocking the government’s “supply-chain risk” designation and ordering agencies to restore ties.
  The ruling cites free speech protections, reinforcing limits on AI use in surveillance and weapons.

• Google Launches Gemini 3.1 Flash Live for Voice AI
  Google introduces Gemini 3.1 Flash Live, a real-time voice AI model with lower latency and more natural conversations.
  It delivers faster responses, better reasoning, and multilingual voice interactions, enabling more intuitive and reliable AI experiences.

• Anthropic Leak Reveals Powerful New AI Model
  Anthropic confirms a new model after a leak, calling it a “step change” in capabilities and the most powerful AI it has built.
  The model shows advanced reasoning, coding, and cybersecurity abilities, raising concerns about potential misuse and AI-driven cyberattacks.

Too Important to Miss

After Anthropic, Gemini Doubles Down on Easy Chatbot Switching

Netflix Raises Prices Across All Plans by Up to 12.5%

Google launches Lyria 3 Pro music generation model

Mark Zuckerberg texted Elon Musk to offer help with DOGE

Last Week’s Poll Result

• What concerns you most about AI infrastructure moving to space?

  Environmental impact (light pollution, etc.) → 45.45%
  High launch and maintenance costs → 27.27%
  Regulatory and governance issues → 27.27%

• Would you trust an AI agent to work directly on your computer?

  No, that feels too risky → 42.31%
  Maybe, but only with strict approvals → 38.46%
  Yes, that would save me a lot of time → 19.23%

• What worries you most about this kind of attack?

  Compromised admin accounts → 44.44%
  Weaknesses in device management systems → 33.33%
  Disruption to healthcare operations → 22.22%

Feedback

Know someone who may be interested?