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 .
@GergelyOrosz ·
Seeing how darn convenient it is having GrokBot integrated into X (an app I’m used to having on my phone all the time) I’m thinking Slack+Teams became the most valuable human/AI interface for work purposes
Salesforce + Microsoft not even realizing this like they should
@OpenRouter ·
Don't bet your company on just one source of intelligence when everyone else is using many.
Choose all kinds of brilliant
@ArtificialAnlys ·
Ling 3.1 Flash 1019, an open weights model and successor to Ling 3.1 Flash, scores 44 on the Artificial Analysis Intelligence Index v4.3, up from 41
@AntLingAGI has released Ling 3.1 Flash 1019, an update to Ling 3.1 Flash, their reasoning model with 560B total parameters, 25B active parameters, and a 1M token context window. It scores 44 on the Artificial Analysis Intelligence Index v4.3, up from 41 from Ling 3.1 Flash. Like Ling 3.1 Flash, Ling 3.1 Flash 1019 is open weights.
Key results:
➤ Ling 3.1 Flash 1019’s agentic capabilities improve over Ling 3.1 Flash. On AA-Briefcase v1.1, Ling 3.1 Flash 1019’s Elo is 1521, up 124 points from 1397 for Ling 3.1 Flash. Its Elo on GDPval-AA v2.1 is 1653, up from 1617 for its predecessor, an increase of 36 points. On AutomationBench-AA, Ling 3.1 Flash 1019 scores 68%, a 6 point increase from Ling 3.1 Flash’s score of 62%
➤ Ling 3.1 Flash 1019’s makes gains in scientific reasoning compared to its predecessor. Ling 3.1 Flash 1019 scores 43% on Humanity’s Last Exam, a 4-point improvement from 39% for Ling 3.1 Flash. It scores 26% on CritPt, an 8-point increase from 18%.
➤ Ling 3.1 Flash 1019 uses more output tokens per task than Ling 3.1 Flash. Ling 3.1 Flash 1019 uses 145k Output Tokens per Intelligence Index Task, more than 40% higher than Ling 3.1 Flash’s per task usage of 103k output tokens
➤ At $1.49 Cost per Task, Ling 3.1 Flash 1019 is more expensive than Ling 3.1 Flash but scores higher on intelligence. This is primarily driven by the higher number of output tokens used. However, Ling 3.1 Flash 1019 is still less expensive than models such as Mistral Large 4 Preview and GLM 5.3
Additional model details:
➤ Size: 560B total parameters, 25B active
➤ Context window: 1M tokens
➤ Pricing: $0.30 per 1M input tokens and $0.90 per 1M output tokens, with cached input at $0.06 per 1M (80% discount)
➤ License: MIT
➤ Availability: Ling-3.1-flash-1019 is available as an open weights model on Hugging Face, and is accessible through InclusionAI’s first party API
@levelsio ·
Talked with my clanker and based on my nutrition I get about 200mg of magnesium per day
You need to get about 420mg per day, so I'm slightly deficient
About 60% of people are magnesium deficient and symptoms can be anxiety (and in turn bad sleep) because magnesium helps your brain regulate your nervous system and stress hormones like cortisol
I don't have any issues (no anxiety, no depression at all, mostly because regular gym and clean nutrition) but maybe good to take it to become even more chill
So I got Magnesium Glycinate supplements
I'll report back on the benefits! 😊
One thing interesting is yeaaaars ago when I had lots of anxiety (before I started going gym and eating clean) I remember taking magnesium and my anxiety just disappeared, so it def works!
@DeryaTR_ ·
I’ve been waiting for the Singularity for more than two decades! Back then, we thought it might arrive sometime around the middle of this century. Yet here we are, with the Singularity seemingly within reach, and more people are beginning to recognize what I have long believed to be the inevitable destiny of humanity.
The pace of change is accelerating so quickly that our usual sense of time is starting to break down. Developments we expected to take decades are happening in years and soon will happen in months then days, and what once seemed like science fiction is becoming part of our daily reality.
It feels as though the world is bending around an approaching singularity. The closer we get, the more distorted our perception of time becomes. Some people are still thinking in terms of the next 20 or 30 years, while some of us already see more changes coming in the next few. We are living in the same world, yet experiencing it on very different timelines.
That is why everything feels so disorienting. Our institutions, expectations, and assumptions were built for a much slower world. As Super-intelligence accelerates discovery and creates the tools for even faster progress, we are approaching a transformation that will redefine science, society, and even what it means to be human.
I’ve argued that we are approaching an intellectual event horizon beyond which our conventional understanding of the future, based on extrapolating from the past, becomes increasingly meaningless. What once seemed like science fiction is becoming reality before our eyes.
That is why the world feels so strange right now, at least for those who are already AI-pilled. We are witnessing the beginning of perhaps the most profound transformation in human history, and I believe we have barely scratched the surface of what is coming!
@lmoroney ·
Filmmakers keep asking me when AI will help with planning a scene before anything gets shot. Google DeepMind's Genie 4, out today as a research preview, is a world model you can explore and change as you go: type a prompt to shift the weather, move the camera or add a person, and the scene adapts in real time. It now generates its own audio too, including ambient sound and dialogue, and holds a consistent scene for about two minutes, double what Genie 3 managed. Gemini 3.8 Flash runs alongside it, planning steps and checking frames to keep the world coherent. DeepMind is upfront that details still drift in longer sessions and in-world text is often unreadable. Trusted testers have it now, with Google AI Ultra subscribers in Project Genie and early Gemini API partners next.
If previs is part of your process, write down the three shots you always struggle to block, and test exactly those when you get access. 🎬
#AIfilmmaking
Link in Comments
@mxstbr ·
🫡
what plugin should we ship next?
@amasad ·
Designers’ contribution to the codebase.
@emollick ·
The repeated argument that AI companies are talking about existential risks as a way of creating hype & inflating their stock prices has always been odd.
Telling people your product may murder everyone does not, in fact, encourage people to adopt it or reduce regulatory risks.
@thdxr ·
i just don't see where the demand for this is today
is there all this hidden deepseek usage that we're not seeing? whenever we talk to providers it's clear all of their other customers are 100x smaller
@MiaAI_lab ·
Finally letting go of my RTX 5090.
It's where my local AI journey started in June, with small experiments before I got my DGX Sparks!
Will have a new owner by tomorrow.
@elonmusk ·
Madness
@elonmusk ·
😑
@svpino ·
AI is literally saving lives right now:
• UC San Diego's AI sepsis model cut sepsis deaths by 17%.
• In a trial of 100,000+ women in Sweden, AI-supported screening found 29% more cancers.
• In a randomized trial involving 60,893 hospital encounters, an AI warning system reduced the hazard of in-hospital death by 35.6%.
• In a randomized trial of nearly 16,000 patients, AI-powered ECG alerts reduced 90-day mortality from 4.3% to 3.6%.
I've never seen technology this useful arrive this fast, and this is just the beginning.
These guys are doubling down:
You can now have agents working inside hospitals, learning from patients, helping doctors make decisions, and keeping the information secure.
@ryancarson ·
Had a nice dinner with @imjaredz and the @DevinAI @cognition crew last night.
Couldn't live without Devin at @HelloUntangle.
Switch to a great cloud harness.
If you're doing local dev, you're working in the past.
Try it and you'll see the light.
@Voxyz_ai ·
last week, quest was breached.
attackers got into quest’s quanum lsm, the system doctors' offices use to order labs and look up results. one platform accessed thousands of offices. quest confirmed patient records were accessed and still hasn't said how many.
the first question i'd ask any health ai: where does the patient data go?
concurrence launched today, built the opposite of most platforms:
𝗯𝗿𝗶𝗻𝗴 𝘁𝗵𝗲 𝗮𝗴𝗲𝗻𝘁 𝘁𝗼 𝘁𝗵𝗲 𝗱𝗮𝘁𝗮, 𝗻𝗼𝘁 𝘁𝗵𝗲 𝗱𝗮𝘁𝗮 𝘁𝗼 𝘁𝗵𝗲 𝗮𝗴𝗲𝗻𝘁.
the agents are forward-deployed inside each hospital. patient data never leaves that hospital, and nothing is shared between hospitals. each one still learns from every patient it treats.
building anything that touches patient data? copy what concurrence did.
@mitchellh ·
Libghostty progress in the past 6 months. It's amazing that NOTHING IN THIS LIST was true 6 months ago, and now its ALL VERY REAL. I remember bringing up the idea of libghostty 4 years ago and most people saying "but why." I did it anyways. And here we are!
Text reproduced in case its easier to read this way:
As a reminder, libghostty is the highly portable, embeddable library that provides core terminal emulation, render state, and more. I always felt that a free, high qualty, lightweight, reusable terminal emulation core would benefit the developer world more than any single GUI application. And that is playing out!
It is and always was my true purpose with the Ghostty project (I said this in my first ever talk about Ghostty!). I've jokingly said many times that Ghostty GUI is just a glorified tech demo for libghostty. That's not fair though, because GUI is great and there are really talented people working on it.
Since I started the project, Ghostty has always gotten people saying "I don't understand why you'd start a terminal emulator today." Well, the answer is libghostty. It's truly unique and its enabling something special (imo!).
Highlights of libghostty since 6 months ago (none of the below was true 6 months ago):
- Libghostty has nearly 100% coverage of all of Ghostty's core (and you can compile with flags to remove any of these features for size/safety): https://libghostty.tip.ghostty.org/
- Libghostty now builds and is tested (in CI) for: Apple (macOS, iOS, visionOS), Windows (MSVC and GNU), Linux, FreeBSD, Android, WebAssembly, and freestanding (i.e. direct on hardware with no OS).
- Libghostty has bindings in pretty much every popular programming language: C, Zig, Go, Swift, Python, Rust, .NET, etc.
- Libghostty's memory requirements have dramatically shrunk to enable the portability above. There's still work to do but a single empty terminal now takes 90% less memory. And at scale I've proven 10,000 active (input/output happening) terminals on a Linux server with 4 cores and 8 GB of RAM.
- Libghostty powers all JetBrains IDE terminals, over 12.5 million active users. https://blog.jetbrains.com/platform/2026/10/jetbrains-termin...
- Libghostty is the most popular core behind the wave of venture-backed "agentic" terminals and editors, such as Herdr, cmux, and many others.
- Libghostty is being discussed and has been verbally accepted to replace the terminal core in Neovim: https://github.com/neovim/neovim/pull/39773 (work still to be done)
- Libghostty powers a really popular and powerful terminal emulator for emacs (Ghostel): https://github.com/dakra/ghostel
- Libghostty's speed is used by Antithesis to fuzz popular terminal programs and has directly led to hundreds of bug fixes sent upstream to programs such as tmux, btop, shells, and other terminal emulators. :)
- And so much more: https://github.com/Uzaaft/awesome-libghostty
Libghostty is really successfully turning into the tide that raises all boats. As it becomes more popular, Ghostty itself gets more performant and stable, and so does everyone else. It is also a vehicle for delivering modern terminal emulation to the widest source possible.
I plan on tagging libghostty 1.0 in the next few months and declaring ABI stability at that point!
@lmoroney ·
China's first permitted fully AI feature opened soft. Hollywood is selling the next ones into AFM. Critics shredded a 7-minute AI Odyssey.
Paying audiences still get the final vote.
Link in Comments
@businessbarista ·
Just spoke to the Head of AI at a huge company.
The #1 thing they're struggling with is... what do I do with the AI laggards at our company?
Like most companies, they’ve found that a Super Pareto Principle is true within their org. 10% of their employees are responsible for 90% of the uplift in productivity/token usage from software engineering to non-technical knowledge work.
They’ve spent a lot of time & resources on AI enablement from LLM foundations to function-specific trainings and skill building. And yet, the power law of productivity increases hasn’t budged.
So now they’re left with a few sobering thoughts and hard questions: 1) 90% of people are overpaid and 10% are underpaid 2) We’d be better off allocating tokens proportionally to the power law, but how do we do that without killing culture? 3) Significant layoffs are likely warranted and wouldn’t impact overall output, but cultural backlash is a big concern? 4) How do we hire more power law people as a non technology company? 5) Is all of our efforts on training and enablement ACTUALLY worth it? 6) How do we not experience a brain drain where the top 10% all leave because they want to be surrounded by other AI-native 10x’ers?
@rileybrown ·
This is sick actually. You guys should stream this somehow.
@AndrewCurran_ ·
Claude is the only author listed on this paper, in contrast to the OpenAI papers, where the human co-authors are listed. There are a lot of strange things about this. It's fun to guess. Conjecture time. I shall reveal my thought traces:
Claude Fable 5.1 and Claude Mythos 5.1 are the same model with the same weights; the only difference is the guardrails. We are about to get Fable 5.5 - in fact, Fable 5.5 is ready to deploy now, and Anthropic could do so at any time. If they follow the same pattern, and I think they will, this implies the existence of a Mythos 5.5. I believe Mythos 5.5 wrote this paper.
The way this is worded -
'Provenance. The core algorithm was found on October 16 by an internal Claude research model with 14M output tokens and no human input. Anthropic researchers worked with Claude to improve the presentation, resulting in this paper. Independently and concurrently, an internal model at OpenAI claims to have proven a similar result [Ope26]'
- makes me think Claude was not prompted to solve this. I think Mythos was asked to autonomously find and choose general open problems to solve itself, as part of an unannounced project to demonstrate recent advances in autonomy and research taste. You can see this taste in Fable 5.5's student; Opus 5.5. This recent advancement in taste and creativity - which happened in September - is not clearly expressed in benchmarks, but it is why so many people are currently switching to Claude as their primary. I think this research project will be announced with Fable 5.5, as part of Anthropic's big media blitz to boost the IPO.
One of the things X has taught me is that many people, if they can't see something moving, can't imagine it moving. This causes them to be continuously surprised by developments in this industry, because the frontier is invisible. I think this project has been running autonomously for some time and has made an unknown, and probably significant, number of discoveries, and not only in math.
Do not think Anthropic has been sitting idly while OpenAI repeatedly makes these huge science announcements. This is a war. This ETH result was released without any announcement from Anth at all despite its significance, and they only did so to respond to OpenAI's ETH reveal. They posted it late last night without saying anything. It was up for hours before levent even mentioned it. What value has one coin to a dragon who sleeps upon a mountain of them? This makes me think they are sitting on a pile of discoveries, a golden hoard, and they are saving them up for the big IPO shock-and-awe campaign which will soon begin. And when it does, they will announce multiple discoveries across many fields.
@RhysSullivan ·
This is great, Answer Overflow has served its purpose and once this goes live I’ll set up redirects to the official discord site and shut down the bot
Has been amazing to help millions of people find what they were looking for but it always should been an official feature
@natolambert ·
I think it's vastly underrated how important org chart and team functionality is to building a good LLM. Over 18 months old, and still one of my favorite all time Interconnects posts. It's why we still often see orgs crumble and need to get rebuilt (e.g. Llama/xAI) and why I believe underdogs can still catch up. Having an effective org and team is worth multiple multiples in compute.
I also think there isn't that much secret sauce to building models. Yes, sure there are some tricks, but most of the gains come from consistent execution and using your compute wisely.
@omarsar0 ·
I am still very bullish on local LLMs. We have yet to see its biggest unlock.
I am particularly curious to see what's possible with unlimited inference.
One exciting application is using local models to run always-on proactive agents cheaply.
Great to see Sentient OS pushing for this. It uses a local LLM on your computer to read your context overnight, then keeps what it learns as markdown files you own.
@FireworksAI_HQ ·
Microsoft EVP @jayparikh joins us for a fireside chat at Fireworks Forge, and we have plenty of questions about building and running AI at @Microsoft's scale.
For those joining at home, tune in to watch the Forge keynote live Nov 3: https://bit.ly/forgelivex
@ElevenLabs ·
Every voice agent has to hear a customer over a TV, a barking dog, or the grandstands at a grand prix.
We wanted to try something even louder, so we borrowed @audif1_’s wind tunnel.
Sorry about the noise, @HulkHulkenberg.
@ryancarson ·
What an exciting time to be alive!
@thdxr ·
when working on apis or protocols or schemas i used to put a lot of effort in making them beautiful because human eyes had to consume them
now i care more about explicitness even if its ugly
@goodside ·
Crazy progress over Genie 3. Imagine where this will be in a year.
@svpino ·
This app does a lot of what Siri should do on your Mac:
It builds context from files, messages, notes, emails, and calendar events, and then uses that information to help you write messages and complete tasks in the apps you use.
And it uses a local Gemma 4 E4B model running through LiteRT-LM!