The Computomatix Times

Griffin, first ever at passing video turing test

I took the early preview for a spin, I'm super impressed. Natural flow, very promising technological advancements from prev gen.

48% of folks who tried it thought they talked to a real human. Full duplex-video is here.

Tavus gave me early access to Griffin and i used it as a startup advisor

i pitched it my company. it called it "a legit play" and then told me SF is "a pressure cooker"

honest feedback from a video call with an AI. wild

the numbers are even wilder. NVIDIA ran a full-duplex video benchmark:

> human ground truth: 3.92
> griffin lite: 3.83
> gemini 2.5 + anam: 2.80

0.09 away from a real human

griffin is a humaninteraction model for live video. it sees, hears and reacts while you talk

@tavus @hassaanraza tavus.io/griffin

Introducing Griffin, the first model to pass the video Turing test.

48% of people who talked to it live thought it was a real human. Previous systems have had a pass rate <3%. It is #1 on NVIDIA's benchmark for full-duplex AI video.

It’s the first Human Interaction Model (HIM).

This is smart and I hadn’t thought about it this way.

Meta may not need to put ads inside Muse... If Muse browses, shops, or makes reservations for you, it creates intent signals Meta can monetize later through ads on Instagram and Facebook.

Velez is one of the best CEOs out there today

This is a great profile about him and the Nubank story

Claude Sonnet 5.5 can (finally) generate motion graphics

15 motion graphics in Claude Code (my 6 steps):

1. Show it an example.
Screenshot a graphic you like (the one on this post works), attach it and write: "Make this move. One HTML file, inline SVG, no libraries, a 6 second loop." Start there and fix it round by round. Say the loop length up front so it does not guess.

2. Write the states, in order.
A reference gets you the look. For the motion, say what happens and when. My command palette brief: the key chip presses at 0.6s, the palette springs open at 0.95s, four letters land between 1.35s and 2.1s and the list filters after each one, the highlight drops a row, Enter at 3.9s, a toast rises. Six lines of plain English.

3. Make time the only input.
Tell it every element is a pure function of t, from 0 to 6 seconds. No timers, no CSS keyframes, and frame 0 equals frame 6. The loop closes itself and any frame can be drawn on demand. The palette opening is a single line: k = spring(t - 0.95, 240, 18).

4. Give it colours, fonts and rules for motion.
Hex codes and font names. Mine are Claude orange #D97757, navy #0F1B2E and DM Sans. Words like clean and modern do nothing. Springs for anything that lands, eases for anything that leaves. Mix them up and it feels cheap. When you can, morph one shape into the next. In my terminal loop, a log line's highlight bar grows into a dashboard card. Same rectangle, new size.

5. Make it check itself.
"Render 8 moments across the loop side by side and fix what's wrong." That is how a black sphere and two cards stuck in one slot got caught. Point at fixes by timestamp: "at 2.0s the second row overlaps the first." Read the sheet yourself too.

6. Export a GIF that stays sharp.
Save every frame as a PNG, then run ffmpeg with palettegen stats_mode=full and paletteuse dither=none. I use 15 frames a second. One palette for the whole clip and no dithering, so flat colours stay flat. Post it as an image, not a video, and check the animation survived the upload.

The 15 in the image include a prompt bar that turns into a web page, a funnel you can watch leak, a donut that unrolls into a bar and a command palette that filters as you type. Each one is a single HTML file with no libraries.

I write about builds like this at aibyaakash.com:
aibyaakash.com?utm_source=ag-20261001

Pick one motion. Write the six lines. Ship it today.

Marty Cagan: "Fundamentally good product work is about thinking. I did not appreciate the lengths that people would go to in order to avoid thinking."

Introducing @Supertake Memory.

Earlier this week, we launched Supertake and shared our vision to equip everyone with the superpowers to invest in their own insights through frontier-level intelligence.

We believe everyone deserves their own personal financial agent, tightly aligned with their interests and working around the clock with the most powerful technology.

To do that right, Supertake needs to know you: your preferences, your beliefs, your intuitions, your history. The same things a world-class financial manager would learn, but only after many years.

Memory is a step in that direction. Supertake now remembers what you tell it, in every chat: your goals, your preferences, your limits, the calls you've made and why, even the nicknames you give your takes.

It carries all of that into every conversation, whether you're building a new take or checking on one, and it keeps track of where each conversation left off, so you don't have to repeat yourself. You also stay in control. Ask what it remembers, correct it, or tell it to forget, anytime.

With today's release, Supertake not only acts upon your insights, but begins to compound them.

Available for all users, right now.

Longtime @stratechery reader, first-time talker. It was a real treat to nerd out on all things Amazon, Walmart, e-commerce--and of course, Muse and the personal agent craze--with @benthompson this week.

stratechery.com/2026/an-interview-with-jason-d…

We also discussed the life and death of Recode, and my conviction that my new publication @ReadTheAisle will be the last job of my career.

Hope you take a read, and then consider subscribing here: readtheaisle.com/join

Google personal assistant is geared at families, called CC

Great name!

labs.google.com/cc/

No one is more equipped to build this than Ben.

He sees it from both sides: the merchant with a store in his head, and the dev who knows what it takes to get it into the code. That’s the uncommon lens Canvas needs.

The store you can picture is the store you can build. Can’t wait to see what everyone makes.

Fox passed on Ted because Seth MacFarlane wanted a $65 million budget and they thought that was too much for a first-time director. Universal said yes, the movie made $549 million, and 14 years later they're launching the fourth product built on that bear.

The interesting part is the format choice. Ted's whole gag in 2012 was a photoreal teddy bear standing in the real world. MacFarlane performed him in a motion capture suit, Wahlberg acted his scenes against empty air, and every frame with the bear ran through a VFX pipeline. The live-action Peacock series carried that same cost structure and still broke the record as Peacock's most-watched original.

Animation deletes that cost entirely.

Drawn, Ted costs the same as every human on screen. The A-list cast trades on-set shooting schedules for voice booth sessions they can knock out between other projects. Drawings never age either, which is why Family Guy has been running since 1999 with the same characters. MacFarlane even brought in Rough Draft, the studio behind Futurama, to make it.

In 2012, Ted was the most expensive character in his own movie. Animated, he's the cheapest.

This is the endgame for every hit franchise. Live action proves the demand, then the IP migrates to the one format that can run for 20 years at flat cost. Peacock just turned its biggest original into a show that never has to end.

2004 was the last year an American teenager could goof off in a school hallway with zero chance a stranger would ever see it.

Facebook had launched that February, locked to Harvard students. YouTube was still a year away. The first iPhone was three years out, and nobody carried a camera that talked to the internet.

So when a camcorder came out, kids hammed for it, because the tape had exactly one destination. A drawer in somebody's parents' house. Maybe a screening at graduation, then twenty years of dust.

Watch the body language with that in mind. Nobody is checking how they read to an invisible audience. Nobody resets their face between takes. The camera was a toy, not a distribution channel.

Researchers have a name for what came after, context collapse. Every audience you've ever had, all watching the same clip at once. Teenagers now run that calculation before they move. The kids on this tape never learned it, and you can see the difference in a single frame.

Low-rise jeans come back every few years. A camera with nothing behind it never will.

so many people underestimate how much easier something is to sell when there’s a good story behind it.

“Mastering storytelling is how you stop selling, and start subconsciously persuading people to do things you want them to do.”

this is a MUST read if you sell anything

RELEASE THE PICS

I love muse.ai. Been using it for shopping, life management, reminder to buy John Mayer tickets and save money by helping me monetize my rewards programs and cut out duplicate and hidden costs on my credit cards.
What companies do people think get disrupted or benefit?

The only problem with my live fact checking bot for zoom calls is that it fact checks me as well and calls out when I am wrong... probably need to fix that :)

We are pumped to announce a long term partnership with @Acquisition_Lab - they’re the go-to place for acquisition entrepreneurs to learn and grow, and a key part of the ETA ecosystem. We’re excited to have them as presenting sponsor!