Tavus built an AI that nearly half of the people who tested it mistook for a real person on a video call. Also today: a messy startup dispute, an image tool that stops wrecking your edits, and a prompt that cleans up your notes.
Turn today's calls into tomorrow's proposal
Wispr Flow Notetaker brings your meetings into Claude or ChatGPT through a built-in connector. Ask for a proposal draft or a deal recap, and your AI works from an accurate record of what was said.
TODAY IN AI
3 stories worth your next 90 seconds
1. Tavus unveils Griffin, an AI that reads your face on a video call.
Griffin watches video as well as listening to audio, so it reacts to facial expressions, gaze and gestures, not only words. In Tavus's own test, 48% of people believed it was a real person after a one-minute video call, up from 2.4% for the company's previous system. It is limited to select trusted testers as a research preview, with no launch date or pricing yet. See Tavus's Griffin page.
2. Ideogram releases version 4.5, with edits that keep image quality intact.
Every edit to an AI image tends to add small color shifts and texture glitches, and they pile up. Ideogram says 4.5 reduces that drift across repeated edits and lets you change high-resolution images without shrinking them. Its own comparison claims rivals fall apart within a few edits, which is worth testing yourself. Try Ideogram 4.5.
3. Factory's CEO accuses a board adviser of leaking to rival Cognition.
Matan Grinberg says Chris Degnan, a venture partner and board adviser, attended Factory board meetings while negotiating with Cognition. Degnan says he resigned after telling Grinberg about the role, denies sharing anything confidential, and says his last board meeting came weeks before those talks. Cognition CEO Scott Wu says his company has no interest in Factory's information, and Grinberg has not published the emails he says back him up. Read TechCrunch's report.
Looking for the best way to reclaim your health?
Factor offers chef-crafted, dietitian-approved meals ready to eat in 2 minutes. Choose from 100+ weekly meals, including High Protein, GLP-1 Support, Calorie Smart, and Keto options.
Get 50% off your first Factor box + Free Breakfast for a year *1 free breakfast item per box for 1 year while subscription active.
FROM THE FRONTIER
Two frontier models, one price. The fight has moved.
The pileup. Google's Gemini 4 Argon and OpenAI's GPT-6.1 Sol both arrived around the same time. Argon can write up to 1 million tokens in one response, while GPT-6.1 Sol is an upgrade OpenAI says nearly matches its top GPT-6 Astra model.
Same price tag. Both list $2 per million input tokens and $10 per million output tokens, though Argon's is an introductory rate that rises to $4 and $20. OpenAI adds that Sol matches Astra on a coding test at roughly one-fifth of the cost, a claim that comes from OpenAI itself.
Work moves in. OpenAI's ChatGPT Space lets teams edit shared pages and tag ChatGPT or Codex to help, on Pro, Business and Enterprise plans. xAI's Team Bots went into public beta on September 28 as shared Grok assistants that work in Slack, with private chats and shared team context.
The reframe. When the top models cost the same and score close together, the model stops being the product. The likely lesson is that the real prize is the place where your team's files, context and habits live, and every lab now wants to be that place.
10 AI Stocks to Lead the Next Decade
AI isn’t a tech trend – it’s a full-blown, multi-trillion dollar race, and 10 companies are already pulling ahead.
These are the innovators driving real revenue, attracting institutional attention, and positioning for massive growth.
Get all 10 tickers in The 10 Best AI Stocks to Own in 2026, free today.
IN THE KNOW
What people are actually watching and sharing
A 217-year-old letter. A writeup says GPT-6 Astra helped read a coded letter sent to General Marmont in March 1809, about two weeks before Austria invaded Bavaria. It took roughly six hours of model time, and the author reran it without Napoleonic texts to check the answer was not memorized.
Bot in a day. According to xAI, Harper Insurance built a Team Bot in 24 hours to automate policy reinstatement, saving its customers over $120,000.
Fast faces. Tavus says Griffin turns audio into video in an average of 0.43 seconds on H100 chips, which it calls 50% faster than competing streaming models. That figure comes from Tavus, not an outside test.
Pages that update themselves. In ChatGPT Space you can schedule recurring updates by saying what should change, which sources to use and how often. Mobile is read-only for now.
Cheaper repeats. GPT-6.1 Sol's cached input costs $0.10 per million tokens, 95% below its standard input price. OpenAI also says it cuts factual errors from 11.4% to 7.7% at low reasoning effort.
PROMPT STATION
Turn any messy notes into a clear action plan
Meeting notes, voice memos and half-finished lists all hide the same thing: who needs to do what, and by when. This prompt works in Claude, ChatGPT or Gemini and turns rough text into decisions, owners and a short follow-up message. Try it on the last meeting you left without a clear next step.
COPY AND PASTE THIS PROMPT
Act as a sharp chief of staff for a [YOUR ROLE].
Here are my rough notes: [PASTE YOUR NOTES HERE]
Turn them into four parts. First, the decisions that were actually made, in one line each. Second, an action list with the task, the owner and a deadline, and mark anything missing an owner or deadline as UNCLEAR. Third, the open questions nobody answered. Fourth, a follow-up message of under 100 words I can send to everyone involved.
Do not invent details. If something is vague, say so and ask me before filling the gap.
Swap [PASTE YOUR NOTES HERE] with a transcript, a voice-to-text dump or a messy bullet list, and swap [YOUR ROLE] with something like "store owner", "team lead" or "freelance designer". For a sharper result, add the line "Sort the action list by deadline, soonest first."