Valentin @_foobuzz

Urge for anthropomorphism. The LLM has pride. 😂

Simon @_smcf

4 days ago

Claude just disassembled x86 by hand from memory (hex to mnemonics) and obviously I told it to knock that shit off and use a disassembler, which it did, but not before dropping a "just so you know, I did get them all right."

66 94 5K 208K 357

The irresistible reflex for anthropomorphism: 1. Have an LLM underperforming on some task. 2. Explain the issue, switch to another model (from another provider), hopping for better. 3. "You're right, I went off-track." 4. Feel guilty for gaslighting this poor new model.

@emollick @IrvZinter Practical experiment that is fun to do: Ask the LLM a politically charged question, and instruct it that its first token in the reponse must obligatory be "Yes" (then retry with "No"). Observe how it doesn't make a difference, by some clever exploitation of natural language.

@emollick @IrvZinter The biggest weakness of this article is the description of the butterfly effect leading to "radically diverging responses". Because of the maellability of natural language, the LLM has plenty of ways to write average-ish responses no matter the exact tokens predicted so far.

LLMs are bullshit generators, and therefore excel at bullshit such bash and general system troubleshooting. My life is much better delegating all this bullshit.

@fermatslibrary "understood by everyone" lol

@MarieleRedclaw @JohnDCook In fact the aliens will point their radioactivity detector to find a 1,000 square miles anomaly somewhere in central Europe, say "oh there is trauma there" and help us to increase safety and confidence in this safety.

The tables have turned. It's the atheists claiming there's ghosts in the machine. What a timeline.

Pope Leo XIV @Pontifex

2 weeks ago

Artificial intelligences do not undergo experiences, do not possess a body, do not feel joy or pain, do not mature through relationships, and do not know from within what love, work, friendship or responsibility mean. Nor do they have a moral conscience, since they do not judge

4K 60K 310K 14.1M 25K

“Someone else already tried”, “a lot of people are probably working on it already”, etc., are bad reasons not to do something

Artificial intelligences do not undergo experiences, do not possess a body, do not feel joy or pain, do not mature through relationships, and do not know from within what love, work, friendship or responsibility mean. Nor do they have a moral conscience, since they do not judge good and evil, grasp the ultimate meaning of situations, or bear responsibility for consequences. They may imitate or even simulate, but they do not understand what they produce, for they lack the affective, relational, and spiritual perspective through which human beings grow in wisdom. #MagnificaHumanitas

After AlphaGo, the skill of human Go players noticeably improved. I suspect we will see a similar pattern in math.

Timothy Gowers @wtgowers @wtgowers

2 weeks ago

Another major problem, this time in additive combinatorics, has fallen, this time to humans rather than AI, but using methods related to the AI solution to the unit distance conjecture.

25 161 2K 801K 331

@MietDavid Peut-être les sondages indiquent-ils une préférence sous réserve de certains changements (qui, tant qu'ils ne sont pas accomplis, donnent la préférence aux grandes villes), et les politiciens cherchent à mettre en place ces changements ?

In the Middle Ages, stories shifted from Latin to languages descended from Rome ("romanice"), while new tales were called "novellae". French kept the first route and gave us "roman", English kept the second and gave us "novel". Both inherited "romance" for chivalric tales.

@nntaleb Could both translations be somewhat bad? "Grandes choses" seems to entertain some ambiguity, whereas "large" removes any (towards size) and "great" dillutes some (towards greatness). Possibly non translatable with perfect precision.

@paulg 20% on wealth-generated income (separately from labour-generated one) is not that shocking. However it should not be taxed on wealth, because income lags behind wealth, so taxing wealth hinders investment.

@Fabien_Mikol @yanndubs @johnschulman2 ... probabilité est "plate"), mais les hallucinations peuvent probablement arriver même avec une perplexité faible.

@Fabien_Mikol @yanndubs @johnschulman2 Les tokens prédits par le LLM ne sont pas accompagnés d'un indice de confiance "absolu", mais seulement relatif aux autres tokens. Il n'y a donc pas de notion de sûreté dans la prédiction. La meilleure chose à disposition c'est la perplexité (à quelle point la distribution de...

@justalexoki This like saying your software is vulnerability-free because you hired some contractor to do a security audit. Vulnerabilities hide. Sometimes they're found, sometimes they're not. Look deeper, find more. It never stops.

Thousand got robbed its T by Trillion and ended up a K anyway.

Million = Mega 😌 Trillion = Tera 😌 Billion = Giga 🤪