sofie n ✿ retweeted
Getting open models production-ready is mostly software tuning. @DebarshiRaha talks it through with @VictorSuOrtiz and the vLLM team at Open Intelligence Summit on Oct 13.
Getting open models production-ready takes constant software tuning.
Exciting to dive into this with Victor Su Ortiz (@MiniMax_AI ) & the @vllm_project team at Open Intelligence Summit by @digitalocean on Oct 13.
If you serve open models in prod, or build AI-native applications, come join us: do.co/4ApUvzV
4
3
4
4,077
Dragonfly GP @HadickM on why personal agents like "Instinct" have no moat:
"I think these products have no lock-in whatsoever."
"Those users are mostly power users here in Silicon Valley, and are mostly the types of people who are willing to experiment. They're also the types of people who are willing to go use a new chat agent, a new model, with the drop of a hat."
"Noah went on a different podcast last week. He talked about, 'Oh, we're doing a billion dollars of volume very quickly.' Well, a billion dollars of volume is actually quite small in terms of the ability to make money."
"The second Muse dropped, you could tell that Muse, because they had better access to compute, was operating more quickly."
"My expectation is that, over time, people will just move from agent to agent, and they will move from whatever and however things are working the best."
"I actually think agents inside your financial applications will be much more important than agents like these in the ability to capture value."
@dragonfly
On this topic, playing around with Muse this weekend and asking it and Instinct to do the same tasks, and Instinct is noticeably (sometimes SIGNIFICANTLY) slower and runs into more problems for the same or similar tasks - clearly compute constrained relatively, which is unsurprising. With more competition incoming, it's pretty clear that these types of consumer first agents are going to require other lock in that isn't just "great experience and some context" and the willingness to burn capital to monetize elsewhere will be needed.
3
3
53
12,169
sofie n ✿ retweeted
Dell Technologies, JERA and RHAELM have signed an MOU to advance national-scale AI infrastructure development in Japan.
By bringing together energy, infrastructure and technology expertise, the collaboration will help accelerate AI computing capacity and support Japan's growing AI economy. del.ly/6013BGveEU
7
12
23
4,846
sofie n ✿ retweeted
Tucker appreciation post! DYK: Hippos can open their mouths an impressive 150 degrees!
7
72
1,027
11,640
The more data you have to look at the better you can use it to optimize a harness, thanks to nemo relay we had a lot more to work with!
Your agent got the right answer. How much work did it take to get there?
We worked with @NousResearch/@Teknium on a hands-on walkthrough of NVIDIA NeMo Relay to collect traces for Hermes Agent. Run two example scenarios, follow the agent’s calls and retries, and see the full trace in Arize Phoenix.
The blog also looks at how Nous used traces and task results to evaluate fixes across repeated runs.
TechBlog: nvda.ws/47tKxjQ
📽️ by our own @Picomoorhead:
21
21
509
35,875
sofie n ✿ retweeted
Trusted Infrastructure for Super Intelligence
Proud of the @Dell team. The first Vera Rubin NVL72 rack-scale system, fully integrated and shipping in volume. 🚀
Thank you @JensenHuang @nvidia. The pace of this partnership is unlike anything I've seen in 40+ years.
53
84
650
110,763
I appreciate financial skepticism of Silicon Valley more than most, but it's retarded to pin your review of this prospectus on the $4.6B in revenue in 2025. Their run-rate was already double that by the end of the year, reportedly $65B in July, and estimated to be $100B by EOY.
🚨 Anthropic IPO S-1 prospectus LEAKED:
Valuation target: $2 TRILLION
>$4.6 billion in revenue in 2025
>$8.06 billion in operating loss
>$42 billion in net loss
The prospectus warns that nearly 25% of revenue came from just TWO customers, and that many top enterprise clients aren't locked into long-term contracts and could cut spending or walk away anytime
Anthropic plans to spend $518 BILLION on compute commitments ‘in the coming year’
96
29
1,097
188,094
sofie n ✿ retweeted
Harvard cognitive scientist Steven Pinker (@sapinker) on AI's future path.
"It will get better. It'll never be omniscient. It'll never be omnipotent. That's out of comic books out of religion. There are laws of physics. There are limits on information. There's no system that could learn the position and velocity of every particle in the universe. So as long as that is true, there's not going to be an AI system that can do anything or can know everything."
Steven Pinker
69
67
235
235,258
sofie n ✿ retweeted
On June 17, skywatchers enjoyed a rare event: a daytime lunar occultation of Venus.
If you missed it, there will be another opportunity to see Venus disappear behind the Moon tomorrow, Sept. 14, for parts of Asia, Africa, Europe, and western Russia.
🔗 go.nasa.gov/46g537d
ALT Venus is seen as it disappears behind the Moon at the start of the occultation on June 17, 2026, from the Mary W. Jackson NASA Headquarters building in Washington. NASA/Joel Kowsky
29
71
511
21,826
sofie n ✿ retweeted
Stimulating piece on Scottish education in @asteriskmgzn: asteriskmag.com/issues/15/th….
I suspect that discussions of canon, culture, and shared context to return to the fore in the coming years, as many skills, narrowly-construed, are ceded to AI, and as some of the excesses of de-Arnoldization, and the processes that Arendt described in her writing on education, start to recede.
I’m curious on the causal linkage between PISA scores and the reforms he describes.
It feels like a very exciting time to be a theorist of education, and I assume that the best work will come from surprising places.
35
35
359
63,967
