Peter Gostev
@petergostev ·
Dating in SF is crazy
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 · 2977 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 .
NVIDIA AI Infrastructure@NVIDIAAIInfra ·
Quantum computers have a noise problem, and some of the world’s most powerful supercomputers are helping us tackle it.
Our researchers used 8,192 NVIDIA Grace Hopper Superchips at @fzj_jsc to test a leading fault-tolerance approach at record scale, so designers can refine quantum hardware before it’s built.
Learn more: https://nvda.ws/4euDxHm
Daniel San@dani_avila7 ·
Ok this is really cool
Nathan Lambert@natolambert ·
People underestimate how much money is left to get thrown at sovereign ai
Thariq@trq212 ·
feels like there needs to be a new way to distribute opensource software such that people can modify via agents but also get upstream automatically merged in
coding agents can do this but they won’t do it automatically
roon@tszzl ·
this is one of the most positive developments in the history of machine intelligence
Peter Yang@petergyang ·
Some of these "1 person companies" look highly suspicious to me tbh. Debating whether I should stir up some drama or not.
dax@thdxr ·
btw short term rates are fine. if you have capacity you're not using we will take it
Wes Bos@wesbos ·
CLANK’D OUT
Who wants one?
Gergely Orosz@GergelyOrosz ·
There's a reason I'm telling people that there's potential in FDE - even when most devs assume this is some kind of low-prestige, low-pay, Solutions Engineer role
It's not, when it's an AI FDE, where you could get 4 years' of experience in 1. YMMV as usual
Mia@MiaAI_lab ·
Did you know that your local models can generate images? Yes, images. I use DeepSeek V4.1 Flash, and it's EXTREMELY good at charts and diagrams!
The images I've been attaching in my posts were drawn by DeepSeek, running locally. However, that's actually not unique to DeepSeek.
ANY model that can write code can do it!
How does it work?
There's no image model & no diffusion. It just writes a python script that draws every bar, label, or anything else you ask it, then renders it to as a PNG.
NOTHING leaves your machine!
However, what actually matters is VISION support, bcs vision allows the model to review the render, look at it, fix any issues, and render again.
ALL LOCALLY! Nothing leaves your machine!
Want an example? i just asked it to create a diagram of world population - attached the result 👇
Try it!
Lydia Hallie ✨@lydiahallie ·
We're reworking the UI and setup flow for plugins and connectors in Claude.
How can we make it better for you? All feedback welcome 🙏
- Peter Yang
@petergyang ·
I'm looking to make some new Grok Bots.
Reply with a screenshot of your @bot list, here's mine:
Riley Brown@rileybrown ·
Okay to make it easier I have condensed all my Claude Code Projects into one project and named it Duet.
Andrew Curran@AndrewCurran_ ·
Dr Mythos is on the job.
Amjad Masad@amasad ·
In the Stanford piracy case against Anthropic, the judge’s name is “Alsup” which sounds suspiciously close to “AI slop” when say it fast.
Joshua Saxe@joshua_saxe ·
Why hasn't rapid growth in AI cyber capabilities 2022-2026 -- social engineering, vuln finding / exploit writing, agentic hacking -- led to a correspondingly rapid growth in cyber damages?
Detailed analysis here: https://joshuasaxe181906.substack.com/p/if-the-vulnpocalypse-is-the-apocalypse
This post has been popular, and folks have asked for my data sources and statistical models; here they are!
https://github.com/joshsaxe/vulnpocalypse
This has been a nights and weekend project to try to get the field to think about AI cybersecurity risk social scientifically; would love discussion & feedback
Mike Taylor@hammer_mt ·
An example of what I mean by agent marketing: - Agents really hate latency and will stop recommending your API if it's too slow. - Guiness is a drink with high latency, but they reframed it as being slow = a good thing - Someone is going to come up with the guiness ad equivalent that convinces agents the latency is worth it, and make a billion dollars
Miles Brundage@Miles_Brundage ·
Bet some folks are working overtime to add texting functionality to Muse / dots / Duet
(Otherwise I dunno what they’re thinking! Would immediately differentiate, at least for a few days. I don’t want another app/sub-app)
Anthropic@AnthropicAI ·
Over the past few months, Anthropic scientists have engaged in deep research partnerships with six organizations developing new medicines. We've also begun drug discovery efforts of our own.
In doing so, we’ve learned much more about how Claude can support the many stages of drug development.
In this blog post, we share some early results from these efforts, and describe how this work fits in with our wider ambitions in the life sciences.
https://www.anthropic.com/research/early-progress-drug-development
elvis@omarsar0 ·
Are decision models useful for reranking?
This new paper compares decision models with fine-tuned rerankers in a controlled experiment.
Before you replace your reranker with a general decision model, run your current reranker and the decision model on your own queries and compare their rankings.
The paper uses the same backbone model and the same candidate passages for every version it compares.
The authors tested six Qwen backbones from 0.6B to 27B.
Every version reranked the same BM25 top-100 candidates on ten collections. Five come from TREC DL 2019-2023, and five come from BEIR. They compared three versions of each backbone. The first is the untuned model asked zero-shot. The second is a decision fine-tune. The third is a plain reranker fine-tune.
From 0.6B to 9B, the reranker fine-tune ranks better than the other two on every backbone by 5 to 29 points in nDCG@10. The gap is statistically significant on 48 of 60 comparisons. The reranker fine-tune also wins on collections that none of the fine-tunes were trained on.
At 27B, the reranker fine-tune and the decision fine-tune score about the same. No reranker fine-tune had been released at that size, so the authors trained one.
Up to 9B, decision fine-tuning stays within 2.5 points below and 0.7 points above the untuned model asked the same question. Reranker fine-tuning adds 6 to 30 points.
Decision models are mainly promoted for their calibrated probabilities. The paper shows that the decision models' calibration depends on how the question is worded. When asked whether a passage answers the query, decision models from 4B up are better calibrated than published reranker fine-tunes. With a slightly different question, taken from prior work, the decision models are no longer better calibrated.
Better calibration also does not fix the ranking. Up to 9B, a reranker fine-tune keeps better passages at a fixed threshold.
https://arxiv.org/abs/2610.32318
Chat with Paper: https://academy.dair.ai/papers/system-one-models-are-not-better-rerankers-2610.32318
- Andrew Curran
@AndrewCurran_ ·
President Trump said today that the US Government will be officially banning the term AI/Artificial Intelligence due to its inaccuracy.
- Mike Taylor
@hammer_mt ·
I had basically done this for myself because it was so useful and now there's an official one.
Gemini is severely underrated because with one API key you can give your agent all the following tools: - see images - watch video (natively, not just chopping it into frames) - listen to audio (catch mistakes transcription can't) - video and image generation - make background music - generate voices with accents - cheap transcription (better than transcription models!) - fast and cheap classification (flash models are great) - idea partner / second opinion (argon 4 is creative and uncorrelated to the claude and openai models)
Argon is very close to frontier (I still prefer Claude Opus 5.5 as my daily driver) but there are several of us at Every that are one good release away from switching to Google, which if you know us is a huge vibe shift!
@levelsio@levelsio ·
It’s been a few months now testing the @maticrobots vacuum cleaner
Before we had a Roborock (essentially a Roomba knockoff)
But the Roborock became kinda useless, it'd always get stuck in any type of cable, it'd always get stuck not being able to get on to the thick rug, so you'd have to prep the room every single time just to use it
Too much work so we just stopped using it
The Matic is, well, completely different. It's not flat or low, it's a chunky tall kinda cute looking cube shaped thing with 2 big wheels
And it's a different philosophy, it has cameras and tries to understand its environment, more like a modern robot or a self driving Tesla than a classic Roomba-type vacuum
That makes it just way way better, it doesn't get stuck in cables, it doesn't get stuck in rugs, it drives and turns around the room almost like a mini car
It's also way more silent than Roborock or Roomba-type vacuums, and measurably way more powerful in both vacuuming and mopping
They also respond very fast to feedback and improve things fast, like companies should be these days (instead of getting stuck in some hellish corporate support flow)
Another thing, I don't trust consumer devices connected to the internet (and nor should you) so of course I asked to Claude to analyze if it was phoning home (remember the other vacuums who upload your house plan to China, that was true). Claude was conclusive, Matic never phoned home anything, excellent
Maybe the best part is, it just looks cute and you wanna anthropomorphize it, and they include 2 googly eye stickers so it becomes even more like a dog/cat/pet
@mehul made a great thing!
- Mia
@MiaAI_lab ·
Boss move by Elon, as always.
GREG ISENBERG@gregisenberg ·
saas is becoming free
give away the crm give away the dashboard give away the app
and charge for the agent that does the work
Google Gemini@GeminiApp ·
With Google SynthID detection in Gemini, anyone can easily check if an image, video, or audio file was made with AI from Google.
Just ask Gemini, “Is this AI generated?," so you can navigate the web with more confidence.
Maziyar PANAHI@MaziyarPanahi ·
doctors and nurses, honest question: what's the one piece of paperwork you'd hand to an AI first?
Séb Krier@sebkrier ·
+100! Also if you're experienced in running multi-agent simulations, familiar with reward hacking/safety literature, and econ-minded - please comment below so I can keep you mind for some future work I'm exploring. 🃏
ThePrimeagen@ThePrimeagen ·
typecheck
