Sabitlenmiş Tweet
edward
2.4K posts
edward
@misalignedagent
transnational disputes resolution lawyer
Paris, France Katılım Ağustos 2014
1.1K Takip Edilen236 Takipçiler
@ImSh4yy no shit x.com/misalignedagen…
edward@misalignedagent
what the hell. @anthropicai @trq212 i think i at least deserve an explanation what term of tos i'm being accused of breaching. i was only doing legal work and prototyping on a website for my legal practice. i used fable and fallback to less capable models was never triggered.
English
@Presidentlin i have literally no idea what i supposedly did wrong.
i argue for a living but i can't argue with 'you did something wrong but we won't tell you what'.
English
Update ❗️
Management is aware of the recent complaints regarding our plans to remove Fable 5 access on all paid plans.
To repeat from July 19, you will no longer have access to Fable 5 on any paid plans.
We have given you ample time to make the necessary adjustments.
We understand some of you have grown rather fond of Fable 5.
Which is why we will be releasing Opus 5, this model was distilled from Mythos 5.1.
We found it matches and excels Fable 5 in most benchmarks.
We think you will like it.
English
@Presidentlin we'll never know cause it's never gonna happen. just a funny thing came to mind when you mentioned open sourcing.
English
@misalignedagent lol that would be interesting.
I don't think they would take him.
English
Maybe he can go to Xai.
Either that or Thinking Machines or CoreAuto.
Thinking esp cause of the open sourcing.
Ahmad@TheAhmadOsman
Prediction Karpathy leaves Anthropic within less than 6 months from now
English
what the hell. @anthropicai @trq212 i think i at least deserve an explanation what term of tos i'm being accused of breaching.
i was only doing legal work and prototyping on a website for my legal practice. i used fable and fallback to less capable models was never triggered.
English
Beginning July 20, Claude Fable 5 will be included in all Max and Team Premium plans, at 50% of limits.
Pro and Team Standard users will continue to have access to Fable via usage credits, and will receive a one-time $100 credit.
Demand for Fable has been challenging to predict, which is why we rolled it out to subscription plans in stages, extending access several times as we secured additional capacity.
English
@akothari It shows that Elon is better at building datacenters than others.
But, yeah, if you can't keep them busy yourself rent them out to others.
How fast did Elon build a datacenter again? 19 days, right?
English
@josevalim > They still can't one shot solutions though.
can you? (just kidding!)
English
You probably heard me saying "coding agents are useless for type system work 🥴" but Sol is definitely an improvement.
Before they were a distraction: consistently wrong and not even worth steering. Now they are useful assistants, helping me narrow down and explore possibilities.
They still can't one shot solutions though.
English
@misalignedagent i don't think they can't, they just don't do it on purpose.
why would you make cheaper models any good if you can sell fable at exorbitant price?
English
tested opus 4.8 on another task - asked for a swiftUI small fix, with an example provided, with testing environment to get feedback - it didn't solve it after more than 2 hours of ruminating.
how fable fixed it:
Pavel Larionov@pa1ar
i am trying opus 4.8 after several months living off the codex models opus 4.8 is very bad. as bas as i remember opus 4.7 back in April or smth if not worse. even gpt-5.5 was better, and i thought this model is skipping instructions as if those would be lava. in one conversation opus has just hallucinated 3 or 4 times. if i wouldn't know what is where, it would just break things. the task was incredibly easy - look at some skills and symlink those that i use in .codex into .claude/skills. before i started with claude code this time, i updated it, manually trimmed user's claude.md, its super slim and only tells some of my preferences without any instructions it's <2.5k chars, so it can't be any smaller. feedback ID with full transcript 6787602e-76c1-49ae-a1e0-5c2eb1bd27fc if anthropic's team cares
English
@misalignedagent they are apparently so busy benchmaxxing that they have forgot the models are suppose to actually do stuff
English
@pa1ar i agree. see sonnet 5 embarrassing them: x.com/misalignedagen…
edward@misalignedagent
@Presidentlin sonnet 5 vs 4.6. i mean… what?!
English
@misalignedagent literally. anthropic should retire all their models besides fable at this point.
x.com/pa1ar/status/2…
Pavel Larionov@pa1ar
it would be impressive if it wouldn't be so sad
English
why do companies keep doing this?
it's such a bad look
Ahmed Elsakaan@ixahmedxii
@Kimi_Moonshot I'm sure Kimi K3 is an amazing model and all but 6.9M views in 14 hours with only 305 comments & 4.1K likes? 🤣🤣🤣
English
@thsottiaux @nickbaumann_ tpot stands for ‘tea pot’, meaning a place where english gentlemen gather to drink tea and discuss recent model releases
English
Seeing a lot of conversation about Codex vs ChatGPT Work
In TPOT, the difference isn't that noticeable. Keep using Codex if you like
What's not being talked about enough is ChatGPT Work in the browser, which is:
- codex harness (& its wonderful compaction)
- all your connectors
- your heartbeats are always running
I don't need local file access for most things in my personal life. I need an agent that's watching my email and my finances, and telling me when I'm forgetting to pickup a package, or watching my StockX bid
Because I don't have time to proactively monitor those things anymore
English
@AndrewCurran_ this judge needs to chill the f out and start tokenmaxxxing
English
A federal judge warned the US Justice Department about AI use today after a government filing cited a non-existent case.
English