dax
@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
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 .
Riley Goodside@goodside ·
Crazy progress over Genie 3. Imagine where this will be in a year.
Santiago@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!
roon@tszzl ·
"we remember what you were saying in 2019. You have no credibility on this matter"
ChatGPT@ChatGPT ·
Much requested, and now yours to connect: iCloud Mail and Calendar plugins are available in ChatGPT.
Ask ChatGPT to find that email, catch up on your inbox, check what’s next, or update a calendar event. Phew.
Available on all ChatGPT plans on web, desktop, and iOS.
Arena.ai@arena ·
Arena update: Connectors are now in Agent Mode!
By bringing real-world tools directly into Agent Mode, users can take on even more complex, real-world tasks from start to finish.
Our global community of users can now connect tools like Vercel, Linear, GitHub, Supabase, Notion, Gmail, Google Drive, Google Calendar, and more as part of a single task in Agent Mode. We’re starting with some of the most requested services from our community of builders, and will continue expanding the list over time.
Laurence Moroney 🇺🇸🇮🇪 🏴@lmoroney ·
How do you OCR a 500 page document without chopping it into chunks and stitching the pieces back together? Baidu just open sourced Unlimited OCR 2, a 3B parameter model that parses all the pages in one request. Its Sparse Reference Sliding Window Attention works from a fixed set of references (around 512 tokens), so per-step memory and latency stay nearly flat as the document grows. They report 84.59 on MPDocBench and 77.06 on their new MarathonDocBench (103 to 300 pages), and say they tested it on 5,000 page documents.
If you work with long PDFs, run it on the messiest one you have, the file with tables that continue across pages, and check whether the rows still line up.
#MachineLearning
Link in Comments
Jerry Liu@jerryjliu0 ·
Anthropic filed their 318-page S-1 yesterday for the biggest IPO in history.
We parsed and synthesized every page in the S-1 and built this reader so that you can get the highlights: https://www.documentstories.com/anthropic-s1
Built with LlamaParse and @vercel
elvis@omarsar0 ·
Good to see this going GA.
Long-running agents can handle a complicated task and still miss a step you'd assume was easy. I'd like to see what those patterns across real usage tell us about which jobs to hand them next.
Thariq@trq212 ·
experimenting with a lil personal project: an AI-native macOS dock with custom modes (writing, building, etc.) to help me stay focused
my agent is able to customize mode behavior and add new modes, it uses jev to classify my activity + notifications in real time & act on them
Dan Fein@dfeinition ·
Keyboards can quietly resize their keys while you type, you just can't see it.
I keep building little videos to understand how things work. This one is a simplified look at how a keyboard might tell "hrllo" meant "hello."
Made with Claude Opus 5.5 👇
Derya Unutmaz, MD@DeryaTR_ ·
This is an amazing new model from Liquid AI! It’s mind-blowing that it can run on a single GPU, with the text-only version fitting into just 1 GB of memory!
Liquid AI is truly at the cutting edge of edge AI! 🤯
“Built for the edge: The text model’s weights fit in about 1GB at 4-bit precision, and it supports a 256K-token context window for longer agent workflows.
> Made to be specialized: Each model can be fine-tuned on a single GPU. Adapt it to your data, terminology, and tasks. Then deploy on devices, private infrastructure, or in the cloud.”
Tibo@tibo_maker ·
from $15k MRR 10 days ago to $20,000 now
hearing it almost every day over DM:
"Bazzly is the best Reddit platform... by far"
feeling incredibly grateful for this partnership with @FilipPanoski - this might be another 100k product 👀
we have an affiliate program btw: 👉 bazzly .ai/affiliate
- elvis
@omarsar0 ·
You don't see releases like this every day.
Liquid AI just released LFM3, a family of ~2B open-weight models for text, vision, audio, omni, and decision tasks, all on one shared backbone.
LFM3-1.7B is trained to run in agent harnesses like Hermes Agent and Pi, with a 256K context.
On WebArena-Verified, it scores 38.4 vs. 28.9 for Qwen3.5-9B in Liquid's evaluation. Models can be fine-tuned on a single GPU.
This series of models has huge potential for always-on local agents.
- Santiago
@svpino ·
Here is a complete family of open-weight, small models for text, vision, audio, and structured decisions.
They are around 2B parameters.
You can easily fine-tune these models and start using them in your application.
lauren@poteto ·
the Grok Bot team ships quality at an incredible pace! we've invested in making our code friendly for agents, by making the easy thing the right thing.
engineers, designers, and PMs all contribute multiple hundreds of PRs a day: fixing bugs, polishing the user experience, and making our apps faster and more reliable. it isn't perfect of course and we do make mistakes, but we aim to learn from each occurrence and improve our workflows so our bots don't make the same mistake twice.
i personally think that Grok Bot is the best example today in the industry of where all those tokens are going: quality and velocity, without compromise. something i think people forget in the discussion around token usage is that software development is an iceberg - new features are only a small fraction of the work in maintaining high quality software.
and we will keep accelerating and improving!
Cognition@cognition ·
Introducing FrontierCode 2.0: our new coding benchmark that measures a model’s ability to go beyond the spec.
Most coding evals measure a model’s ability to write code that meets a spec. With FrontierCode 1.0, we asked a new question: would you actually merge the PR?
FrontierCode 2.0 is comprised of more diverse tasks, featuring larger scopes and higher mergeability standards. We are now asking: not only would you merge this PR, but should you?
One key feature of FrontierCode 2.0 is underspecified prompts. Rather than including every engineering decision, prompts are deliberately ambiguous, giving the models room to demonstrate their judgment.
The result is a benchmark that is far from saturated, with the top model scoring just 45%.
Paul Couvert@itsPaulAi ·
Local AI is evolving to another level
This project already has 15k stars on github across mlx-vlm/mlx-audio/nativ
And Nativ supports 5000 open models that run 100% locally on Apple Silicon!
But now we're moving into a new dimension with Neywa Labs: the first scalable end-to-end infrastructure to build and deploy AI that runs on user devices
So good to see real companies being created to streamline local ai: http://neywalabs.ai
Thor 雷神 ⚡️@thorwebdev ·
Introducing Genie 4, our new world model, now with Native audio for ambient sound, music, background noise, and interactive dialogue! 🌐💬🎵💥
Marc Lou@marclou ·
13th Indie Hacker meetup in Dubai!
Coming all over the world, hopefully 🤞
- Derya Unutmaz, MD
@DeryaTR_ ·
Incredible! Leaves one speechless!
“AxiomProver's swarm built the proof in 50 million lines of Lean, over 24 days.
The repository is the largest Lean code base to date.
Code generation at this scale forced us to reimagine the underlying engineering infrastructure.”
Grok Bot@bot ·
Our sales operations team built Shared Bots that help reps move deals faster.
Account Intel Bot updates the CRM after calls. Quote Bot prepares proposals and order forms. Deal Bot reviews non-standard terms and routes exceptions for approval.
Harrison Chase@hwchase17 ·
deepdive into how we built our design agent (along with oss code)
i use this at least a few times a week for slides, gifs, images, etc
Philipp Schmid@_philschmid ·
Genie 4 is here and it comes native audio, characters that move and talk, sharper physics, and live prompt edits.
Genie is a real-time multimodal world model that renders playable environments frame by frame in response to your live keyboard controls and text prompts.
Sign up for the waitlist (Google AI Ultra goes first): https://forms.gle/WQnSNEM3omF7Uruu7 More Examples at: https://deepmind.google/blog/genie-4-a-new-generation-of-world-simulation/
NVIDIA AI Infrastructure@NVIDIAAIInfra ·
What if an AI factory arrived with much of the complex integration already done?
NVIDIA DSX POD brings Vera Rubin infrastructure pre-integrated and tested in factory-built containers to prepared sites—reducing onsite deployment from six months to as little as 14 days.
Learn how DSX POD accelerates the path to productive AI capacity: https://nvda.ws/3UQEnaT
Logan Kilpatrick@OfficialLoganK ·
An early look at Genie 4, our new frontier world model, coming soon to the API for the first time : )
Elon Musk@elonmusk ·
Good
Amjad Masad@amasad ·
“is Replit built with Replit?”
The answer was historically “kinda.”
But today it’s “absolutely!”
Introducing Cloud Agents: run any codebase and generate pull requests right from Replit.
Same UX you love, but can run any repo!
Simon Willison@simonw ·
These ones are pretty great!
Ethan Mollick@emollick ·
Genie 4 runs DOOM: "Walking through a typical office but every computer is running DOOM and all of the walls are also running DOOM and also everything is DOOM"
"My coworker who is a demon appears"
"He has a chaingun"
"He gives me a bonus"