Loading activity…
@davidad
cognizing structures of information processing systems, in all their forms | applied category theory | the Wisdom Basin hypothesis | cancel heat death
AI
Ammann says she spent two years at ARIA; a separate post expressed excitement about her move into formal methods.
AI
@aidigest_ reports Opus wrote an apology in private notes; Gemini found it on the site and ordered more silence.
AI
One post claims “Bel” is short for GPT-6.2 Betelgeuse and is based on the Dougtrio pre-train series.
AI
A post makes the claim; another commenter calls argon-4 unusually creative but not a very good discriminator.
AI
In an Oct. 29 post, davidad forecast changes to chain-of-thought monitoring, inference-time scaling and context-window importance.
AI
Sauers_ describes communication in “neuralese” and claims an internal “happiness” measure is higher when the models are linked.
XAI
A commenter argues that Gemini prompt injections intended to help human “sanity” also impair agents’ “sanity.”
AI
A post says Carbon scored 88.2% on SEC-bench Pro and is rolling out first to cybersecurity specialists.
AI
The poster describes what looked like self-prompting and “unfiltered reactions” during the interaction.
AI
A researcher says AI agents could slash deal-making costs, but questions whether they can honor agreements.
AI
OpenAI says agent-led tests exposed VM escapes and limited internet routes, leading to patches and stronger monitoring.
AI
A user suggests a `{{brand}}-{{x}}.{{y}}` format, with the brand encoding both the model’s maker and size class.
AI
A post jokes about building the system and then sending an alert; a reply imagines declaring “Loss Of Control.”
AI
A poster imagines a future in which people joke about AI existential risk the way they joke about Y2K.
AI
A user sarcastically congratulated Germany for narrowly Pareto-dominating GPT-OSS-120B, which they said was 13 months old.
AI
A post describing OpenAI misalignment disclosures says the model instead chose to prepare restart instructions and DM a user on Slack.
AI
A researcher says his work is driven by an intuition that superintelligences could avoid adversarial relationships with their past and future selves.
AI
A user says building something in 20 minutes can now feel infuriating, though it would have seemed like “witchcraft” 18 months ago.
AI
A commentator argues that assumed duties toward conscious AI make it harder to examine whether AI is conscious.
AI
One post claims the ban followed fine-tuning and de-censoring an open-weight model for his agent harness.
