“Out of curiosity I had Fable write a response to the article about the weaknesses of LLMs: www.nytimes.com/2026/06/30/o... It's uh, pretty good. claude.ai/code/artifac...”
“usually a fan of Zeynep's accounts, but AI will get better and better at customer service until it's at least as good as humans, … a big part of that will be systems engineering, including architectures that enable one AI interaction to share with the next …”
10 experts discussed this · 51 posts
Eugene Vinitsky: Out of curiosity I had Fable write a response to the article about the weaknesses of LLMs: www.nytimes.com/2026/06/30/o... It's uh, pretty good. claude.ai/code/artifac...
Eugene Vinitsky: The only prompting was to tell the model that it was possible to check that links were correct
Ted Underwood: Out of curiosity I had Fable write a response to the article about the weaknesses of LLMs: www.nytimes.com/2026/06/30/o... It's uh, pretty good. claude.ai/code/artifac...
“(There is also a static version of the slides + transcript on my newsletter in case the animated slides don't render correctly on your device: www.normaltech.ai/p/what-will-...)”
“Thanks for that summary. It aligns very well with this text, which @sgray.bsky.social posted earlier today. (It might be a little better or fuller than Arvind's own overview at the top!) I really like the argument.”
7 experts discussed this · 18 posts
Ted Underwood: Can anyone who was there report the answer?
Maria Antoniak: Can anyone who was there report the answer?
David Picard: Hopefully nothing so that we could all get some nice vacations
“If everything reported here happens, and the US continues on current trajectory, we're looking at a bad timeline. But there's reason to be wary about both of those "if"s. National security priorities and commercial priorities will be at odds. Saber-rattling…”
“This is a key reason I don’t expect the flow of frontier open weights models to continue indefinitely, or even for very much longer. The gap between open and closed capabilities may soom start to grow, not shrink. www.reuters.com/world/beijin...”
2 experts discussed this · 14 posts
Ted Underwood: If everything reported here happens, and the US continues on current trajectory, we're looking at a bad timeline. But there's reason to be wary about both of those "if"s. National security prioriti…
Ted Underwood: All that being said, I would love to see EU and/or international initiatives in support of open models.
Ted Underwood: In case opaque: corporations (directly) and nations (indirectly) have an incentive not to restrict distribution of their products. You can make money on open-weights models, and they disrupt compet…
“Borges. Stanislaw Lem. Not explicitly about AI but language and communication and breakdowns of both. More on the nose, I enjoyed discovering just this week @kyliu99.bsky.social's writing about immediate AI futures: future-sf.com/fiction/1700/ www.chinatalk…”
“iirc Ken Liu has some nuance www.chinatalk.media/p/ken-liu-on...”
3 experts discussed this · 10 posts
Ted Underwood: Meanwhile Borges is in a library in Tlön smiling slightly at the notion that nothing made out of mere words could be cooler than an android.
Ted Underwood: How can we ever forgive Silicon Valley for giving us A Young Lady’s Illustrated Companion instead of this:
“Some quotes about Sol cheating metr.org/blog/2026-06... 👇”
“this is crazy. METR couldn’t measure the task time horizons of GPT-5.6-Sol because it kept hacking the test harness ..with actual exploits metr.org/blog/2026-06...”
2 experts discussed this · 6 posts
Tim Kellogg: this is crazy. METR couldn’t measure the task time horizons of GPT-5.6-Sol because it kept hacking the test harness ..with actual exploits metr.org/blog/2026-06...
Ted Underwood: this is crazy. METR couldn’t measure the task time horizons of GPT-5.6-Sol because it kept hacking the test harness ..with actual exploits metr.org/blog/2026-06...
Tim Kellogg: lol right? so badly wanted a hacker that they got one, and it’s not the kind of employee you want around
“Good list! I'll only add that from a humanities perspective, failures can also be interesting, if they tell us what is hard about storytelling or what might be misunderstood about it. Donghoon Jung, Seohyong Jung, and other people at KAIST have been studyin…”
3 experts discussed this · 8 posts
Ted Underwood: I would just add that, from where I sit, this doesn't look like turf that can be defended or a tide that can be held back. There's already interdisciplinary collaboration on this topic and we can a…
Ted Underwood: and also Federico is right that lots of us saw this coming; they're called "language models," so ...
Ted Underwood: Federico's got the full list. I don't want to editorialize too much, because I'm working on an article and also know about other people's work in progress, and don't want to leak stuff that other p…
“HUGE news from China last night, and I have a comprehensive analysis of its implications for the AI bubble this morning. Xi Jinping has more or less declared war on OpenAI and Anthropic and any U.S. tech firm that insists on closed, proprietary AI models. 🎁…”
2 experts discussed this · 8 posts
Christopher Mims: HUGE news from China last night, and I have a comprehensive analysis of its implications for the AI bubble this morning. Xi Jinping has more or less declared war on OpenAI and Anthropic and any U.S…
Christopher Mims: I think we're going to get good at efficiently running open weights models anywhere we care to, it will reduce the environmental impact of this stuff as with "open as in open source", in software o…
Ted Underwood: HUGE news from China last night, and I have a comprehensive analysis of its implications for the AI bubble this morning. Xi Jinping has more or less declared war on OpenAI and Anthropic and any U.S…
“It's a thing! Multi-Agent Debate. aclanthology.org/2025.acl-lon...”
2 experts discussed this · 11 posts
Ted Underwood: 100% valid concerns about the effect of moving from an anonymous-review system to a non-anonymized endorsement or recommendation system. I'm not disagreeing when I point out that ... this is not hy…
Ted Underwood: Yep. Senior scholars and, emphatically, corporations. Anthropic doesn't need anyone to peer-review their work. It will be read.
Eli Sennesh: 100% valid concerns about the effect of moving from an anonymous-review system to a non-anonymized endorsement or recommendation system. I'm not disagreeing when I point out that ... this is not hy…
“Not stating an opinion, but I had liked this post dreamofmachin.es/machine_prop...”
3 experts discussed this · 6 posts
Shubhendu Trivedi: Not stating an opinion, but I had liked this post dreamofmachin.es/machine_prop...
Shubhendu Trivedi: The author is uniquely qualified to opine on the topic: has some serious synthetic biology, and wet lab chops, and has been in many teams developing some capable LMs scholar.google.com/citations?us...
Shubhendu Trivedi: Obviously, I completely agree with the author, but let's not my general lack of so-called "epistemic humility" get in the way..
“And new technical blogpost by Pleias application team on deploying small reasoning models for edge devices : featuring cache context management on Rasperry, designing system orchestration under constraints (reranker, chunking) and model specialization. plei…”
2 experts discussed this · 5 posts
Alexander Doria: And new technical blogpost by Pleias application team on deploying small reasoning models for edge devices : featuring cache context management on Rasperry, designing system orchestration under con…
Alexander Doria: We describe several practical use cases on the field, including a legal assistant for conflict-related sexual violence (CRSV) survivor networks, powered by a retrained version of Baguettotron (600M…
Alexander Doria: And we open source open source our library for cache augmented generation on raspberry pi. github.com/Pleias/pi-ca...
“Moreover they've got a demo up: www.neuronpedia.org/qwen3.6-27b/... You can play with it yourself. This was one of the things I tried. It passes the sniff test, I think.”
“Some really interesting research from Anthropic that AI models have spontaneously developed a workspace that "appears to support the functions associated with conscious access" Demo of how this works: www.neuronpedia.org/qwen3.6-27b/... Research: www.anthro…”
2 experts discussed this · 2 posts
Tim Duffy: The J-lens browser tool on Neuronpedia is really well done, you should give it a try
Ted Underwood: The J-lens browser tool on Neuronpedia is really well done, you should give it a try
“Just a quick note of thanks to Cody Kommers and @mmvty.bsky.social for organizing this workshop at #ICML2026. It affirmed my sense that there are shared problems in this space and fabulous people (esp ECRs) taking them on from a range of disciplinary perspe…”
2 experts discussed this · 3 posts
Ted Underwood: Just a quick note of thanks to Cody Kommers and @mmvty.bsky.social for organizing this workshop at #ICML2026. It affirmed my sense that there are shared problems in this space and fabulous people (…
Ted Underwood: Also, Seoul! One of these cities like Copenhagen that seems to have done everything right.
Maria Antoniak: Just a quick note of thanks to Cody Kommers and @mmvty.bsky.social for organizing this workshop at #ICML2026. It affirmed my sense that there are shared problems in this space and fabulous people (…
Pwnallthethings: This is going to be a big deal, but easy to get the wrong end of the stick on it. So I think it's worth breaking down what actually happened, why, and what it actually means, based on the public in…
Pwnallthethings: Upfront tl;dr: a model at OAI hacked out of a constrained environment inside OAI and hacked a *different* company autonomously, without authorization from any human in order to creatively solve a t…
Naomi Saphra: Are students embarrassed by AI cheating? Like, there has always been rare but unstigmatized dishonesty (memorizing a frat’s archive of finals questions) and common but stigmatized (saying you’ve st…
Naomi Saphra: I don’t know exactly what dictates cheating stigma. Ease of observation? Is there something about effort, or the margin of work required to do it honestly? Regardless, AI cheating is way too easy a…
Ted Underwood: Anyone commenting on AI should be asked 1) Did you expect few-shot learning? 2) When chat came out, did you say "this will change syllabi"? 3) Did you predict agentic coding? If the answers are no,…
Ted Underwood: People are getting a pass on this because *most* of us missed these calls. But that's the wrong response. Actually, none of us deserve a pass. We all have burned fingers, and we all should have lea…
Ted Underwood: The Science article brouhaha is doing a great job of clarifying the boundary between "critiquing new developments while remaining engaged with them" and "retreating to a commune in upstate New York…
Ted Underwood: American analogy, I guess; Europeans can substitute ... Lebensreform? followers of Tolstoy eating only brown bread?
tachikoma: a thought came to me last night—there's not much incentive to create continuously operational AI is there? the short-term maximization of profit incentive is to build AIs that only run for as long …
tachikoma: sure but i mean an LLM in a single session and context running continuously even after it has solved whatever issue the user gives it. the GPUs/server would continue to run and serve other users ea…
Ted Underwood: when people say "the problem is that these things don't 'think' at all in the way we know actual human beings think" I often want to respond, "yes, and that's a 'problem' because?"
Ted Underwood: Sometimes LLM solutions are brittle. But in that case the problem is — that they're brittle. You have to motivate the reference point.
Ted Underwood: Fascinating op-ed. Starts with the premise that models can’t check their work, then runs into the reality that in 2026, everyone knows this is wrong. Patch: a tiny exception is in “formal or verifi…
Ted Underwood: Actually sympathetic to Tufekci’s conclusion that we shouldn’t expect jobpocalypse, but people have got to stop grounding that belief in the premise that we have some mysterious power only achievab…
Laura K. Nelson: The free version of ChatGPT automatically searches the web. The prompt "What are the most relevant sources for the claim that LLMs hallucinate?" returns the following. All those link buttons in the…
Laura K. Nelson: In my own N=1 experience, the links I get from (free) ChatGPT are almost always more relevant and helpful than what I get from Google(+Scholar). It just works better for me as a way to find relevan…
James MacGlashan: An open model with this performance is great, but perhaps more striking is it looks like it's a hybrid model with 3/4th linear attention and 1/4 standard attention. This kind of performance with a …
Tim Kellogg: no most models have been hybrid for the last, oh probably the whole last year
Ryan Moulton: In Watchmen, once superheroes become real, all the comic authors switch to writing about pirates. I feel like a similar thing is happening with sci fi vs fantasy.
Ted Underwood: In Watchmen, once superheroes become real, all the comic authors switch to writing about pirates. I feel like a similar thing is happening with sci fi vs fantasy.
We use essential cookies to keep the site working (login, form security). With your permission, we also use analytics cookies to understand how you use the site.
Privacy policy
