jpicone @me.jamespicone.name · 05/10/2026Medivo from Jazz Jackrabbit. I was something like 6 and i insisted that mum finish the level for me because it was too scary. 100
jpicone @me.jamespicone.name · 24/09/2026honestly this is not much of a hack, and bypassing anti-bot stuff is imo desirable behaviour. 270
jpicone @me.jamespicone.name · 15/09/2026The great thing about the king's indian defence in rapid is that someone will blunder at some point, the last blunder is the only one that matters, and you're prepared for chaotic bullshit. 000
jpicone @me.jamespicone.name · 15/09/2026No worries, it wasn't much effort. I was hoping our interlocutor would let us know if it was right or not because I found it in a Relevant Way. 030
jpicone @me.jamespicone.name · 12/09/2026I'm slightly irked by the coverage of the UN's map resolution because things keep saying that it makes Africa larger when Mercator has ≥1 areal scale everywhere so an equal-area projection actually makes Africa smaller. (Quote from www.theguardian.com/world/2026/s... ) 🧵1/2 110
jpicone @me.jamespicone.name · 07/09/2026Anyway erdös problem 1196 was solved by a random student just asking off the shelf gpt5.4 chatgpt.com/share/69dd1c... for the chat and www.erdosproblems.com/forum/thread... for the problem. This was probably an easy erdös problem, but it is novel mathematical reasoning. 🧵2/2 120
jpicone @me.jamespicone.name · 07/09/2026Why didn't you post the tweet where he says it's gpt 5.6? The one directly before that one? Here's the chat chatgpt.com/share/6a60b2... 130
jpicone @me.jamespicone.name · 07/09/2026Actual banner I ran into near my parents house with the same vibes (seventh day adventists) 000
jpicone @me.jamespicone.name · 07/09/2026The original rant also explicitly claimed that LLMs "only have success" on a specific class of tasks so stepping back to "oh this isn't empirical" is a pretty obvious attempt to dodge that they were dead wrong. 🧵1/2 160
jpicone @me.jamespicone.name · 02/09/2026A real problem with AI discourse is that anti-ai people are sometimes too ignorant to understand why something an LLM did is impressive. So you tell them that a model improved lower bounds on the Riemann hypothesis and they think it's sort of like calculating pi to a few more decimal places. 🧵1/2 280
jpicone @me.jamespicone.name · 02/09/2026From this story www.abc.net.au/news/2026-09... . I had to go to the page for the associated campaign and then scroll horizontally through their weirdly-presented FAQs to find out what they actually mean, which seems to be opting out of personalised recommendation. God knows why I would want that. 110
jpicone @me.jamespicone.name · 02/09/2026I weep for the cs 101 lecturers that have to undo this confusion. 2417
jpicone @me.jamespicone.name · 02/09/2026Mate I'm linking you to chatbots finding counterexamples to 30-year-old conjectures with minimal direction. Every time you've tried to engage with that empirical evidence you've made a fool out of yourself. The guy just asked for a counterexample and then encouraged the model a bit! 100
jpicone @me.jamespicone.name · 11/08/2026Another Climbing Coverage Out Of Context moment. This was in slow-mo in the original broadcast; all i've done is clip it and drop the sound. 010
jpicone @me.jamespicone.name · 06/08/2026I haven't seen the movie, but the plot summary on wikipedia makes me not super into this pick :P. 010
jpicone @me.jamespicone.name · 03/08/2026My bank is releasing a new online banking system and deprecating their old one and the password requirements for the new system almost make me want to change banks. No whitespace? Wtf? What are you doing with the password that makes that necessary??? 010
jpicone @me.jamespicone.name · 25/07/2026is it a good sign when the support page for a brand looks like this? 100
jpicone @me.jamespicone.name · 19/07/2026Was watching a light show in a park and a possum decided to just hop up on a tree next to me. Wild. Really not very bothered by people or a bunch of robo-swans. 010
![Claude prompt: Any idea what study, if any, this moron is referencing? [Screenshot of bisk making the inaccuracy claim]
Claude output:
Tracking down the source behinda specific ha... >
Almost certainly Dahl, Magesh, Suzgun & Ho, "Large Legal Fictions: Profiling Legal Hallucinations in Large Language Models" (Stanford RegLab/HAI, arXiv Jan 2024, published in Journal of Legal Analysis 16, 2024). It found hallucination rates between 58% (GPT-4) and 88% (Llama 2) on specific, verifiable questions about randomly selected federal court cases. The Stanford press release led with "69% to 88%"- the range excluding GPT-4– which is what most secondary coverage repeated, and which rounds to his "70-90%". Credence ~0.85
that's the source. arXiv) Stanford Law School](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:om3vjisekjaxk2xsyirswxfu/bafkreifuqoguzivnxwhe44xwpaegv3z6uczypgbm7i7y2e3u5fcl5fuweq)
![...As we discuss in $5, LMs are not performing natural language understanding (NLU), and only have success in tasks that can be approached by manipulating linguistic form [14]. Focusing on state-of-the-art results on leaderboards without encouraging deeper understanding of the mechanism by which they are achieved can cause misleading results as shown](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:om3vjisekjaxk2xsyirswxfu/bafkreifxlft5yzxawzzqrp4svjljbs6sbjervuswn7adgtfqaoy2we6nry)
![When we perform risk/benefit analyses of language technology, we must keep in mind how the risks and benefits are distributed, because they do not accrue to the same people. On the one hand, it is well documented in the literature on environmental racism that the negative effects of climate change are reaching and impacting the world's most marginalized communities first [1, 27]. Is it fair or just to ask, for example, that the residents of the Maldives (likely to be underwater by 2100 [6]) or the 800,000 people in Sudan affected](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:om3vjisekjaxk2xsyirswxfu/bafkreieuvdso7iqyr7f5wbr4ytr3uourjelmabqowppv2q55udhj4yigrq)
![by drastic floods' pay the environmental price of training and deploying ever larger English LMs, when similar large-scale models aren't being produced for Dhivehi or Sudanese Arabie?8
And, while some language technology is genuinely designed to benefit marginalized communities [17, 101], most language technol- ogy is built to serve the needs of those who already have the most privilege in society. Consider, for example, who is likely to both have the financial resources to purchase a Google Home, Amazon Alexa or i Annle device with Siri installed and comfortably sneak](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:om3vjisekjaxk2xsyirswxfu/bafkreigvhisyrcrhnhejsh3e43dc4r3zuqag2w52sulbzsi3f3svuxvhsq)
