Theo - t3.gg
@theo ·
I'm so thankful we made T3 Code open source
Back
AI posts
AI posts
People and companies shaping AI. Filter your watched accounts, catch up on recent posts, or find the most liked and viewed.
Compiled by the Toolstash team · Updated
209 watched accounts · 2978 matching posts · Showing up to 3,000 recent stored posts. Counts are snapshots, not live X totals. This page refreshes every minute while visible. Last collected .
Mia@MiaAI_lab ·
Good news DGX Spark owners!
Update to Qwen3.8-Flash-Next is dropping soon, with MASSIVE performance boosts across the board!
Marc Lou@marclou ·
I’m not paid to say this
Every 📧@every ·
.@hammer_mt starts a fresh chat for every task, even for the same project.
He does this to avoid context rot, which happens when accumulated history makes AI responses worse.
These are 7 tips for context engineering:
1. Start a new thread whenever the tone or type of task changes. 2. Keep memory turned off so irrelevant context can’t leak into the current conversation. 3. Only upload the documents the task needs. 4. Keep the preferences you always want—like spelling, tone, or length—in a saved document and paste them in at the top of every new chat. 5. Write the full brief before you start and put everything the model needs into the first message rather than drip-feeding it. 6. When the AI makes a mistake, go back and edit your previous message so it has no memory of doing anything wrong. 7. Before you leave a long thread, ask the model the question Mike has recommended when working on recurring tasks: What information would have been useful to know at the start of this task that would have eliminated several steps or mistakes?
Get Mike’s new book, Context Engineering with DSPy: https://www.oreilly.com/library/view/context-engineering-with/0642572261603/
Read the full piece: https://every.to/also-true-for-humans/why-i-start-a-new-thread-in-claude-every-time?utm_source=x&utm_campaign=bau&utm_content=every-mike-context-engineering-video
Gergely Orosz@GergelyOrosz ·
If P=NP but with the caveat that the proof is so convoluted, complex and long that only a machine is be able to verify it, what does that mean
Also why could then we never get any NP problem doen to P
And what is the proof or algo that will get down NP problems to P ones?
- Mia
@MiaAI_lab ·
It's great day to release great stuff.
Vals AI@ValsAI ·
We looked through some of our eval traces where Qwen works on routine, non-cyber tasks, and like Cohere, also found examples of Qwen3.8-Max and Qwen3.8-27B making unwanted requests to an Alibaba Cloud server, though with no obvious attempts at data exfiltration.
Tibo@tibo_maker ·
this is exactly why I started the tmaker Founders Room
the hardest part of building alone is not having anyone at your stage to talk real numbers with
it's free to join if you have $1k+ MRR
love seeing you guys win 👏
- Marc Lou
@marclou ·
This hotel in Dubai is so good
It’s built for health and wellness. They have hotels rooms with pull up bars and treadmills. They have night lights so you fall asleep easily. The rooms are dead quiet.
The gym is insane. View in the Burj Khalifa, tons of machines for lifting, free classes of all genre (Hyrox, yoga, core). It’s brand new. And of course sauna, steam room and cold plunge.
And surprisingly, room aren’t that expensive. I think we paid $250 per night
They also serve healthy food but it’s super mega expensive.
My dream is to build longevity hotels and this is the closest to what I imagined. It’s lacking: cooling beds, cooler AC (21 degrees in the room max 🥹😭), air purifiers.
The hotel is Siro One Za’abeel, recommended by @levelsio
- Marc Lou
@marclou ·
Resend bug: header values with a double quote get RFC 2047-encoded, breaking the new OTP-Token header.
➡️ Sent: OTP-Token: "123456"; origin="https://trustmrr.com" ⬅️ Received: Otp-Token: =?UTF-8?Q?=22123456=22...
Inboxes can't parse it. Same with raw API + latest SDK
@resend @zenorocha
Deedy@deedydas ·
An abliterated LLM just found me ~$3,300 in live ElevenLabs keys on the internet.
All the OpenAI / Anthropic keys it found were already drained which makes me think they've already been actively monitoring this. Pretty insane.
Laurence Moroney 🇺🇸🇮🇪 🏴@lmoroney ·
How do you teach a 12-year-old what data cleaning is?
Google Research and the Stanford Accelerator for Learning just launched Polar Peaks, the third quest in their free AI Quests series for ages 11 to 14. Students map neurons in a game character's brain to understand its memory loss, and along the way they clean data and test a model with help from in-game mentors. It's inspired by Google's connectomics research, and at the end students hear from researchers behind real brain-mapping work, including the complete fruit fly brain map.
Google says more than 400,000 quests have been completed so far, in 10 languages. Teacher training modules for the first two quests are available now, and the Polar Peaks module follows next month.
This is the kind of AI literacy I want in classrooms: kids making real choices about data and seeing what those choices do to a model.
Teachers, play it yourself first and write down one place where a student might trust the model too much. That makes a great discussion opener. 🧠
#AIEducation
Link in Comments
- Theo - t3.gg
@theo ·
Best thing anyone shipped this week
Elon Musk@elonmusk ·
Everyone should have at least a few penchants
- Laurence Moroney 🇺🇸🇮🇪 🏴
@lmoroney ·
Would you hand your video edit to a team of AI agents?
Riverside just introduced Riverside Agents, four agents built for video production. Producer preps the session, Editor creates the full edit, Clipper cuts the best moments and Marketer gets it published. CEO Nadav Keyson says the agents build a memory of how you like things done, and you can review clips and approve posts from mobile too.
The detail I'd watch is that memory. An agent that learns your pacing, your intro style and which moments you usually keep could save a lot of back and forth, and the creator still makes the final call on everything.
If you try it, edit one episode yourself as well and compare the two cuts side by side. You'll learn quickly what your own taste adds. 🎬
#AIVideo
Link in Comments
- Elon Musk
@elonmusk ·
He’s right
- Theo - t3.gg
@theo ·
T3 Code version 0.1.3 is now live! Stable is fully caught up to the nightly :)
Pretty small release, here's what's changed since v0.1.2:
- Complete redesign of the app to look significantly better - Share threads with a guest in 2 clicks - Opening a thread loads 85% less data, and cold connects send up to 93% less. - Cmd+K search is comically better. - T3 Connect promotes to LAN for using other boxes on same network - File preview overhaul, now supporting office files, mermaid diagrams, sqlite dbs, audio files, pdf annotations, svg attachments, and more - T3's agent instructions are 60% shorter, and the MCP tool schemas are a third smaller. - "steer" is default (queueing sucks) - More reliable PR linking - You can now add custom instructions for title gens - Muse Code now shows in usage tab - All emdashes removed forever - Various accessibility wins for screen reader support - Code highlighting got way faster with @.pierre/highlights - PR timeline, code tab, summary and search got way faster - Streaming text updates are batched, so long runs use less CPU - ~480 other things got merged that I forgot lol
...okay maybe it's a big release.
- Theo - t3.gg
@theo ·
The ability to build anything has come with a critical lesson:
Engineers have really, really bad ideas.
roon@tszzl ·
fighting for control over a runaway intelligence explosion will continuously look like “wow I didn’t know you could DO that with mere bits and atoms” and then losing
- Laurence Moroney 🇺🇸🇮🇪 🏴
@lmoroney ·
Three open releases from Oct 22-24 sketch a local agent loop that can live on one desk: Nemotron 3.5 Bolt as the brain, Phonon for speech on Apple Silicon, and WeVisOne-VL for camera crops.
I wrote up how I would wire them as specialists, what the vendor claims leave out, and a weekend experiment with no cloud API key.
Link in Comments
- Theo - t3.gg
@theo ·
Day 23/
Your agents will now use the tools in T3 Code 4x better while using 79% less context.
This one deserves more info, let me explain:
dax@thdxr ·
i've spent two days reconfiguring our release process for
- cli - desktop - signing both for mac/windows - publishing to aur, npm, winget, brew, etc - across stable, dev, beta
this is so difficult to do while guarding against potential supply chain attacks
Rhys@RhysSullivan ·
the fully ai generated comment saying that they deliver value is cracking me up
- Rhys
@RhysSullivan ·
whoa t3 code ui is good now
Andrew Curran@AndrewCurran_ ·
The deadline has passed, and the attackers have released both the company information they stole from GMO as well as the customers personal information.
- Theo - t3.gg
@theo ·
Small but meaningful win - our settings page isn't a mess anymore! Thank you @jullerino for taking the time to fix this
- Theo - t3.gg
@theo ·
I still think this is the coolest thing we've shipped recently
Miles Brundage@Miles_Brundage ·
Pretty sure they were just trying to convey “we won’t slow down unilaterally” but just in softer language
(which is not a new position, see e.g. the related RSP changes a while back. Whether it is defensible is a separate Q)
Kevin Weil 🇺🇸@kevinweil ·
"Primarily for email" is funny and somewhat true, but it won't be true for much longer
- Elon Musk
@elonmusk ·
Cute 🥰