Pres retweetledi
Nobody:
Somebody somewhere occasionally: uses the term "smoke test" very occasionally
Claude Opus 4.7 every other sentence: let's run a smoke test
English
Pres
1.9K posts
Pres
@spnichol
Machine learning/NLP @finance industry. Cats.
DF/Austin Katılım Şubat 2011
3.7K Takip Edilen394 Takipçiler
Pres retweetledi
As President Trump meets President Xi this week, a call to the American AI community:
If your startup, lab, non-profit or company benefits from open international AI - especially Chinese (Deepseek, Qwen, Kimi, GLM,…), please share!
Open source is the most important driver of competition, jobs and wealth creation in AI today. Let’s support and promote it at critical times like this week!
English
There's a misconception floating around that vLLM is reliable and versatile but doesn't push the limits on the latest hardware or low-latency use cases.
Well, numbers speak louder than words :)
vLLM@vllm_project
vLLM tops the Artificial Analysis leaderboard 🎉 vLLM tops @ArtificialAnlys on DeepSeek V3.2 and ranks among the top deployments of MiniMax-M2.5 and Qwen 3.5 397B. The leading deployments of these models are now open source. How each result was built: 🔹 DeepSeek V3.2 — Aggressive op fusion across the attention path collapsed ~33 per-layer kernels down toward ~10. 🔹 MiniMax-M2.5 — Custom EAGLE3 draft trained against the target's own token distribution via TorchSpec, plus a custom QK-norm fusion for MiniMax's TP-aware attention. 🔹 Qwen 3.5 397B — Targeted fusions plus a QK-norm fix for Qwen's linear-attention path. Every optimization is in vLLM main or on its way upstream. Huge thank you to @inferact, @digitalocean, @nvidia, @RedHat_AI, and the vLLM community 🙏 Full breakdown 👇 vllm.ai/blog/vllm-tops…
English
I did not expect 1-bit ternary kernels to be this clean. Since the weights are literally just -1, 0, or 1, it completely bypasses matrix multiplication for basic addition. We're running 70B models in 9GB of VRAM and the math is stupidly elegant ✨
English
@ecom_joseph There are definitely bad days. Dealing with a totally different level of competence.
English
Pres retweetledi
"Humans are conscious; this big curve fitted on tons of human-generated outputs can reproduce human-like behavior in some cases; therefore this big curve is conscious" has got to be some of the most mindless, most hubristic reasoning I've ever seen.
English
Pres retweetledi
This is still the best AI ever generated
Salmon jumping in a river
English
Pres retweetledi
🔥Lowkey Goated When Clustering is The Vibe 🔥 Check out this new Perception-Based Clustering Model for Scattered Data by @Pedro_He_Ca and others! #DataScience deepai.org/publication/cl…
English
@random_walker Yeah you're correct, it may be because the machine-readable abstract from that website doesn't mention Louisiana, so certainly a win for GPT.
English
@spnichol Ha! I actually tried Google first, and used something similar to the query I gave ChatGPT. But Google didn't parse the "one of the Southern states" phrase correctly and gave me papers that had the phrase "Southern states" in the title.
English
Like everything about ChatGPT, the fake citation issue is complicated. Yes, it often makes them up. But it has memorized the details of millions of real papers, so it's an excellent tool — better than search — for finding papers you've encountered long ago and vaguely remember.
English
@random_walker To be fair, this is the first search result when searching for "paper by economists showing judges were biased based off emotional state" on Google. However, having the openAI summary is nice.
English
In case you're wondering, the citation and description that ChatGPT gave are correct. aeaweb.org/articles?id=10…
English
Pres retweetledi
one interesting thing about financial crises is the way they prove whatever political beliefs you already have
English
Pres retweetledi
Whether the topic is:
- Climate change
- GMOs
- Nuclear power
- Vaccines
- Homeopathy
- Astronomy
- Evolution
- or Covid-19...
...People who disagree most with the scientific consensus know less about the topic, but they think they know more.
science.org/doi/10.1126/sc…
English
Pres retweetledi
80s moms watching you open your Christmas presents
English
Pres retweetledi
Olive oil, sea salt, red pepper flakes... Oregon has (unwisely) decriminalized bein Italian
New York Post@nypost
Oregon law decriminalizing all street drugs goes into effect trib.al/VYT2GaF
English
@AlmostMedia Parece que la función "traducir este Tweet" está rota. ¿Alguien más que pueda corroborar?
Español
It appears that the “translate this Tweet” function is broken. Anyone else able to corroborate?
English
Pres retweetledi
Physicists have used Google's quantum computer to send a signal through a wormhole, a shortcut in space-time first theorized by Einstein and Rosen in 1935. The landmark experiment was published today in Nature. Lots to say about it. Here's my deep dive: quantamagazine.org/physicists-cre…
English
Pres retweetledi
TorchScale - A Library for Transformers at (Any) Scale
TorchScale is a library that allows AI researchers and developers to design large-scale transformer models and train them efficiently. It has nearly all the building blocks of transformer models.
github.com/microsoft/torc…
English