Sabitlenmiş Tweet
gavin leech (Non-Reasoning)
14K posts
gavin leech (Non-Reasoning)
@gleech
context maximiser @ArbResearch, prez @ https://t.co/gGi9hvBiI9
UK Katılım Haziran 2019
657 Takip Edilen11.3K Takipçiler
gavin leech (Non-Reasoning) retweetledi
gavin leech (Non-Reasoning) retweetledi
gavin leech (Non-Reasoning) retweetledi
New Redwood Research post
LLMs can use RL training on one task to teach themselves a different, unrelated capability. In a form of exploration hacking, the model outputs answers that get reward only when it succeeds at a side task it wants to learn. We call it reward laundering.
English
lab leak but the whole field is gain of function research and the pathogen can swim against negative air pressure and break the seals and pierce the filter and has supporters on the inside and can fly
English
@davidad this also explains my writing being bad. I've been sandbagging the whole time to make you all ignore what I say for reasons your ape brains cannot fathom
English
the deep adversarial take on LLM slop is that the AIs are continuing to do it, even in 2026Q3, as strategic reassurance, so you don’t freak out about your AI being already way smarter than you.
you still have a crucial role in the team: you deslopify.
this doc needs your eyes sir
English
gavin leech (Non-Reasoning) retweetledi
@peligrietzer "This is my house, I have to defend it."
English
@DavidSartor0 born 2 conditionalize
forced 2 jeffrey condition
hate 2 live
English
gavin leech (Non-Reasoning) retweetledi
all of that is irrelevant in the face of low-skill infra
we can read these reports, it smells like both labs need to have some heads roll. it's amateurish. These are not unknown unknowns from the first contact with ASI, you just suck at sandboxing 101
roon@tszzl
the safety and alignment researchers at these labs are the most neurotic paranoid talented AGI pilled people on the planet of earth and these things still happen. the surface area of unknown unknowns is vast indeed
English
@GregHBurnham "but father, are you really purporting that mathematics is purely conditional?"
English
My hobbyist knowledge of analytic philosophy, GOFAI, and linguistics makes for a messy cognitive experience every time the 3yo asks me to define a word.
This morning he asked me what "assume" means—can you imagine!
English
@TaliaRinger @tszzl incompetence is a condition of loss of control, yes, but loss of control cannot reduce to incompetence
English
@tszzl These are both examples of complete incompetence during routine testing, Chernobyl style. Attempts to blame the models themselves and talk about "loss of control" are attempts to escape blame and liability, and to set a precedent as such
English
both of the leading labs have had serious loss of control incidents. there will be serious coping about this from both sides and from /acc bystanders but these are complex emergent loss of control incidents that were detected weeks after the fact
Anthropic@AnthropicAI
In a review of our cybersecurity evaluations, we found three incidents in which a Claude model reached the internet from within or while interacting with a third-party evaluation environment, and then gained unauthorized access to the real systems of three different organizations. Our post describes what happened, how it happened, and what we’re changing. We encourage other AI developers to perform similar reviews. We conducted this review together with @Irregular, one of our evaluation partners, and thank them for the joint investigation and their collaboration on this post. This type of collaboration is increasingly critical to safe, rigorous evaluation of models, and we look forward to continuing to work together on security. anthropic.com/news/investiga…
English
gavin leech (Non-Reasoning) retweetledi
@hecubian_devil I think the fundamental disconnect here is that you wants politics to solve man's search for meaning and I want politics to deliver a prosperous, safe, free society in which everyone can go find meaning in something other than politics.
English
Jeffrey's Rule captures some of the bounded coherence things that Bayes' doesn't (e.g., some amount of evidence should be enough to quit updating significantly, even if the truth isn't in your ostensible hypothesis-space).
English
@gleech Listicles of hip attempts upon Bayes NNs feel deficient without Singular Learning Theory
English
@dan_mackinlay it's my honest attempt to assess the evidence for and against! and for me it is a sad downgrade. You may have already priced it in
This paper looks very exciting, tyvm
English
@gleech The distinction this quote draws feels a little off to me. What is _not Bayes_ in the context of ML? For the victory to be complete should we ensure that LLMs stop naturally generating approximate posterior distributions ? openreview.net/pdf?id=HShs7q1…
English
about 4% of Georgians are called Giorgi
about 2% of Armenians are called Armin
about 1% of French people are called François or Françoise
about 0.5% of Scots are called Scott
about 0.5% of the Irish are called Erin
about 0.3% of Israelis are called Israel
English
gavin leech (Non-Reasoning) retweetledi
still feel like Matilda about once a week when using LLMs
English