FeltSteam0

[FeltSteam0 @FeltSteam]

@chaitinsgoose https://t.co/cmT8am6alM

[FeltSteam0 @FeltSteam]

@GaryMarcus @ai_sentience @geoffreyhinton On the Alphafold point I mean like why would we spend our time attempting to investigate the interiority of a sea anemone over studying things that can use language and exhibit complex behaviours like humans

[FeltSteam0 @FeltSteam]

@mad_muppet @edmundheaphy Im not exactly sure what his point was here though. Not that I have read the article yet but why was he relating it to the autocorrect feature? I mean LLMs can obviously be used as autocorrect but they aren't themselves just autocorrect

[FeltSteam0 @FeltSteam]

@mad_muppet @edmundheaphy how do they predict the next token?

[FeltSteam0 @FeltSteam]

@eriskiiii I mean on this AISI evaluation the different between Mythos Preview and Claude Opus 4.6 is the same between Opus 4.6 and Sonnet 4.5 which is a really noticeable gap

[FeltSteam0 @FeltSteam]

@GaryMarcus Maybe he's thinking there will be a point where the models will be intelligent enough for most people then at that point fixate on making that level of intelligence much cheaper as has been the trend

[FeltSteam0 @FeltSteam]

@GaryMarcus @ai_sentience @geoffreyhinton I think it’s realisable to say the same probably applies to AI; Claude is not automatically conscious because of this, but it has far more consciousness relevant markers than AlphaFold, just as a mammal has far more than a sea sponge.

[FeltSteam0 @FeltSteam]

@GaryMarcus @ai_sentience @geoffreyhinton I guess to illustrate we cannot draw a hard line between mammals, fish, insects, worms, anemones, and sponges. Yet we still reasonably prioritise some animals over others based on clusters of evidence.

[FeltSteam0 @FeltSteam]

@harobuilds @deredleritt3r I mean the most optimistic people haven't really changed their "AGI" date. It has been a few years away because 2030 has always been a few years away and still is?

[FeltSteam0 @FeltSteam]

@tszzl How long until we get the Saitama model

[FeltSteam0 @FeltSteam]

@scaling01 Are they going to rerun some of the other models with the updated prompt? I at least wonder how GPT-5.5 would score with the same updated prompt.

[FeltSteam0 @FeltSteam]

@teortaxesTex If we do get some form of AGI/ASI relatively near term it's going to be super fascinating to see how this mixes with interplanetary travel though, im really looking forward to this possibility.

[FeltSteam0 @FeltSteam]

@teortaxesTex We aren't anything close to this for us yet though so it's more of a hope and really only possible with advanced AI's for now at least but if we do get into the depths of the singularity who knows

[FeltSteam0 @FeltSteam]

@256_rc @scaling01 >"We find that the total effective stock of human-generated public text (not even images/video) data is on the order of 300 trillion tokens"
https://t.co/PkV0lMbDBX

[FeltSteam0 @FeltSteam]

@Angaisb_ It would be cool to see a side by side comparison between o3-mini and GPT-5.5 with the same prompt

[FeltSteam0 @FeltSteam]

@thsottiaux https://t.co/iZl5AZi4DB

[FeltSteam0 @FeltSteam]

@JustinBleuel @ChatGPTapp I do have one annoying issue im facing though not in ChatGPT but rather in Codex, it seems that sessions with a large number of images in the context seem to run really slow (with really long "Thinking".. times) and seem to trigger "reconnecting" more often, at least for me.