Maziyar PANAHI
@MaziyarPanahi ·
isn't radiology the first thing AI fully automates? it's images in, a report out
i bet by the end of 2027 it runs almost on its own
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 · 2981 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 .
Arena.ai@arena ·
First impressions of Gemini 4 Carbon’s preview by @GoogleDeepmind are in. @petergostev just put it through a long run of 3D generations, creative prompts, visual explainers and UI tasks, compared against Gemini 4 Argon, Claude Opus 5.5, GPT-6.1 Sol, Muse Wonderful, Kimi and DeepSeek.
Watch now to see an early preview of the latest model coming from @GoogleDeepMind.
0:00 Intro: Gemini 4 Carbon 0:15 First test (White House): Carbon vs Argon 1:28 Carbon vs Claude Opus 5.5 2:42 Carbon vs GPT-6.1 Sol 3:06 Carbon vs Muse Wonderful and Kimi 3:48 Why test with 3D generations 4:45 Playable 3D game: Carbon, Argon, Sol, Opus 6:28 Where Carbon falls short: gaps in detail 7:50 Stonehenge 8:30 Golden Gate Bridge 10:41 Golden Gate Bridge: GPT-6.1 Sol comparison 11:53 Carbon's creativity 15:33 Monet paintings transition 18:37 Van Gogh paintings world 21:47 Why creativity matters in a model 23:30 Another 3D prompt: Carbon vs Opus, Sol, Muse, DeepSeek 26:27 London through the ages 29:56 Istanbul 31:44 Harbin ice city 33:20 New test: visual explainers (how lightning works) 35:34 John Snow and the cholera map 38:20 UI work 39:45 Golden Gate Bridge completion animation 40:43 More UI generations 41:55 Ski resort site and ramen ordering 44:20 Verdict
Lydia Hallie ✨@lydiahallie ·
Dynamic workflows don't have to be hundreds of subagents btw! I use them for any task I want to rerun the same way, even just ~10 subagents
It's a JavaScript file you can edit, so you have full control over the subagents and the order they run in. More predictable than a skill!
Josu San Martin@josusanmartin ·
Given the pricing, it seems that they are increasing the density of the memory, rather than adding more chips.
Vals AI@ValsAI ·
To round out the Vals College Tour, we are wrapping up at the @Princeton Career Fair on Oct 29-30th🐯!
Princeton alumni, @conjfrnk will be there to answer your questions about Vals, post grad decisions, and hand out exclusive merch. Come say hi, we can't wait to meet you.
Higgsfield AI 🧩@higgsfield ·
Mobile game ads never look like the real game.
Now you can turn a viral ad into a full mobile game.
> Drop a gameplay video, a screenshot or a prompt > Build a mobile game with viral mechanics > Turn it into a playable ad on the Enterprise plan
Higgsfield Playable is live now.
ElevenLabs@ElevenLabs ·
New from ElevenLabs research: Today we’re sharing more about Cue, our new forced alignment model. It places every character on the audio timeline with more precision than any other aligner we tested. That means the right words show up at the right time.
Forced alignment is the step that turns a transcript into timestamps. It’s how subtitles land on time, how dubbed lines match the speaker, and how clicking a word in a transcript jumps to the moment it was said. Most open-source aligners are built for short, clean clips in a limited set of languages. Cue is one model that handles recordings of any length in a single pass. On real-world audio, it was more accurate than every open-source aligner we tested in all 29 languages of our benchmark. It aligns an hour of audio in about half a second on one GPU.
Cue already powers every Speech to Text transcription on ElevenLabs, and it’s available on its own through the Forced Alignment API.
Read the full post: https://elevenlabs.io/blog/cue-aligner
Fidji Simo@fidjissimo ·
“Once discovery gets cheap, the hard part is clinical trials and patient data” Couldn’t agree more, and that’s where the bottleneck is about to be. We’re about to see an explosion of novel drug ideas, but without understanding patient biology and how to direct these drugs to the right patient subsets, we won’t materially increase the likelihood of success of clinical trials. That’s why @ChronicleBioAI is building the AI data platform to match the right drugs to the right people based on their observed biology, for every chronic illness.
Theo - t3.gg@theo ·
Happy to help the little guys out here and there 🫡
Alex Lieberman@businessbarista ·
It feels like the internet is breaking.
AI has created an entirely new paradigm in cybersecurity and I want to learn as much as I can about it.
So I found a world-class cyber expert to give me a 60-minute cyber primer on Friday at 12pm ET. And you're invited (link to register below).
David Campbell (@dcam_ai) is former Head of AI Security Research at Scale AI, where he architected and lead some of the most advanced AI red teaming and resilience programs in the world.
Here's what we'll talk about & the questions I'll ask:
1) What cybersecurity has historically meant - Explain cybersecurity to me like I'm a smart CEO who's never had to think about cyber. - Give me the cast: nation-states, ransomware gangs, insiders, hacktivists. Who actually costs companies the most money? - Walk me through the eras of cyber security: castle-and-moat firewalls, then cloud and "identity is the new perimeter," then zero trust. What changed each time? - How was a security team traditionally organized? - What are the key terms/language you need to know to be cyber-literate? Translate SOC, red team, blue team, pen test and CISO for me.
2) How AI changed it - What's the single biggest thing AI changed about cyber, in one sentence? - What did AI do to the types of attacks that are done and the manner in which they’re done? - What did AI do to the cheap, everyday attacks: phishing, deepfake voice calls to finance, fake vendors? - Did AI change who the attacker is? Can a teenager now do what took a nation-state in 2015?
3) What AI-native cybersecurity looks like - You wrote that AI security isn't bullshit, but we're securing the wrong thing, and that prompt injection isn't the real risk surface. Explain. - Companies are handing AI agents keys to email, code, CRM and payments. What new attack surface does that create? - What is AI red teaming, how is it different from a classic pen test, and what does a good engagement look like? - If you were building a security company from scratch today, what would be AI-native about it that incumbents can't copy?
4) What being great at cyber looks like for a business today - I'm a CEO of a 1,000-person company. Give me the five things I must have in place in 2026. - Before I roll out AI agents across the company, what's the security checklist? - How should a board measure security now? What are the 2–3 metrics that matter? - Build, buy or partner? What should a mid-market company never try to do in-house?
Link to register here: https://luma.com/tenexlabs-ztrt
- Arena.ai
@arena ·
Gemini 4 Carbon shifts the Pareto Frontier.
Séb Krier@sebkrier ·
When thinking about the market for frontier capabilities, people would benefit from internalizing some Juran and Gryna (1988)!
Alex Imas@alexolegimas ·
It is a joy to see people making beautiful things.
Andrew Curran@AndrewCurran_ ·
From the blog:
NVIDIA AI Infrastructure@NVIDIAAIInfra ·
The semiconductor industry behind AI’s breakthroughs is now bringing that intelligence into how chips are designed and produced.
See how NVIDIA accelerated computing and AI help the industry move from chip design to quality inspection, using insights from the fab floor to improve models and optimize operations.
Watch the full demo: https://nvda.ws/4iei47W
Vox@Voxyz_ai ·
Like a lot of you, I use Opus 5.5 to write code and Codex Astra to review it.
But Astra has a bad habit of flagging bugs that are technically possible but 𝗮𝗹𝗺𝗼𝘀𝘁 𝗻𝗲𝘃𝗲𝗿 𝗵𝗮𝗽𝗽𝗲𝗻 in practice. Opus sees the logic checks out and fixes it, then Astra flags the next one, and the loop never stops.
Add this line to your review prompt, skill, or AGENTS.md:
"Rate each bug from 1 to 10 on how often it would happen and how bad it would be. If the two scores multiplied are below 5, just file it as an issue and leave the code alone."
Tune the threshold of 5 to your project.
For Opus, the one fixing the code, add one more line to CLAUDE.md or AGENTS.md:
"Do not design for rare or future cases nobody asked about."
Daniel San@dani_avila7 ·
Restrict AI and people will use it anyway, just without oversight
Give them the right tools with Enterprise Settings, Compliance API, and OpenTelemetry, and you get control, observability, and valuable data for training your own models
Worth following @jhan_virani for more on AI in asset management 👇
Every 📧@every ·
This week at Every:
• @RandyCounsman edited an entire Every Podcast episode in Everycut, the video editor built by @kieranklaassen.
• @kieranklaassen crossed 20,000 followers on X.
• @hammer_mt rewrote one headline and took simulated agent citations from 2 out of 240 sessions to 240 out of 240.
• Our ops team wrapped up months of work to get Every collecting sales tax in three more states: New York, Ohio, and Maryland.
• @danshipper personally replied to readers who filled out our survey—and had a post featured by LinkedIn News.
• Yash Poojary and @beckyisj skipped the App Store and built their own apps with Claude.
• The Thesis crew entered the home stretch of preparing our conference on work and AI.
• Grant Harvey joined Every to write the daily edition of Context Window.
• Arielle Shipper hosted a Women in AI night with women from across the team—and one special guest, Leo.
Mitchell Hashimoto@mitchellh ·
Siri is still bad, like really really bad. I opened an email with my Uber Eats confirmation, there was a live activity on the screen with progress, and I pulled down and asked Siri "what did I order for lunch?" And it thought for like 2 minutes and then said it doesn't know. 🤦♂️
Okay so the harness and models are clearly bad. Well, then next I opened up the Siri GUI app on macOS and HOLY. It's the worst looking, beginner, basic bitch app I've ever seen. Did someone spend more than 5 minutes on this?
Whoever worked on this app along with the full chain of people who checked off on this needs to be terminated immediately. This is 0/10 terrible software, so embarrassed for Apple to ship this.
Riley Brown@rileybrown ·
we'll see when we test it. These charts dont matter
Jerry Liu@jerryjliu0 ·
Introducing the Document Parsing Protocol (DPP)
There are hundreds of different OCR models and tools, but they might all parse a document in different ways.
Some errors are very clear-cut (e.g. hallucinating text, or completely misaligning a table), but many errors are much more ambiguous (e.g. how to represent split tables sharing a header). It's an issue that plagues most parse benchmarks too, because they inherently reward a certain style even if other styles might be semantically correct.
We introduce a protocol that proposes a standard way of parsing these documents into a correct format. We've curated a list of 140 rules that cover literally every aspect of doc parsing, for instance: - tables: borderless whitespace tables, header-only tables, horizontal slicing, side-by-side mini tables - forms: key-value regions, checkboxes, checklists, ids and labels, fields outside the form, - charts: standardized chart output, multiple charts, gantt charts, line charts
You can use DPP as an evaluation rubric for LLMs when determining the quality of parsing a document.
It's not a full evaluation benchmark (yet), but an initial evaluation guide, and meant to encourage a discussion on what are the "best" ways to parse a document.
Check it out: https://www.parseprotocol.com/guide/forms-forms-and-key-value-regions
Blog: https://www.llamaindex.ai/blog/introducing-the-document-parsing-protocol
- Andrew Curran
@AndrewCurran_ ·
The Department of Justice is making inquiries into how the NVIDIA Hugging Face deal would affect competition.
Aaron Levie@levie ·
Agents are going to consume 1,000X more tokens than we consume today.
Because they’re going to be running in our workflows, processing all of our data, building out nearly all software, securing enterprises, and much more. To do this at scale we will need a variety of approaches at the frontier of AI.
This includes continuing to push the boundaries of closed models, while also increasing the capability of open weights models, making them more secure, as well as making them more efficient for large workloads.
Very exciting to see what @Suhail is building with Particle on this front.
Mike Taylor@hammer_mt ·
Google bought urchin tracking and renamed it google analytics and made it free. Tt was the best product on the market and completely zero cost, one line of JS to install... ...and yet mixpanel, amplitude, omniture, appsflyer etc all were billion dollar companies. People dont want to trust google to grade their own homework. Same thing with Anthropic - it doesn't matter how good claude managed agents gets, there will always be demand for cross-platform to avoid lock-in
Mia@MiaAI_lab ·
Worth revisiting
@levelsio@levelsio ·
I guess we're omarchyposting now but here's the very funky screensaver
Peter Gostev@petergostev ·
There's one model that I want that I'll never get.
Astra/Fable-based Sydney. Just imagine that.
Derya Unutmaz, MD@DeryaTR_ ·
Unbelievable new AI model Gemini 4 Carbon from @GoogleAI!
It beats Astra and Opus 5.5 on science, math, cybersecurity, and agentic coding! As I wrote before, “don’t bet against Google!”
- @levelsio
@levelsio ·
I figured out Omarchy's keybindings
Action + K
And it gives you all the shortcuts!
- Daniel San
@dani_avila7 ·
Trust me, connect Muse to Todoist and keep your tasks in sync
I know Muse has its own task manager, but I prefer Todoist because it’s simple, supports MCP, and lets me manage the same tasks from other AI agents like Claude Code
One task list, multiple agents
