@adonis_singh World peace is crazy lmao.
For what AI can do about that, GPT3 already had that solved "everyone stop shooting each other", there you go.
English
T. Rebentisch
23 posts
this has to be ragebait
Gary Marcus@GaryMarcus
@midsusnight @skdh World peace, a million things in biology and medicine and climate change, batteries, etc etc etc I am not even saying Astra’s problems aren’t worth solving. I am saying people are overgeneralizing from “exceptional math performance” to “ominpotent” and that’s just plain wrong.
English
The new DeepSeek is even crazier than expected.
At just 284B parameters, it beats GLM 5.2 and comes insanely close to Opus 4.8.
All while being 1/10 of the size and almost 100x cheaper.
I knew innovation was coming, but I did not expect it to hit this hard.
Pro is definitely going to crush Fable and Sol.
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)@teortaxesTex
DeepSeek-V4-Flash is updated. «DSBench-FullStack is an internal full-stack development test set, and DSBench-Hard is an internal Coding Agent hard-problem test set» «The official release of DeepSeek-V4-Pro will follow soon» These scores are… quite a lot better than Preview.
English
C. Kirk suffered an injury in #49ers practice today. Sources say he’s stable.
Please god.
English
@AdamHoltererer lol your „evidence“ are 2 benchmark scores where one of thems is even part of the other. K3 massively outperforms even Sol in multiple relevant domains AND is less annoying to work with.
English
In other words, K3 is in trouble.
OpenRouter@OpenRouter
GPT-5.6 Terra & Luna are 50% off exclusively on OpenRouter!
English
@bridgemindai So you are willing to pray API prices for K3?
English
Kimi K3 open weights are live and it is already running 3x faster.
The weights dropped this morning. Fireworks is already serving it at 45 tokens per second with 1.44s latency.
I have spent 11 days waiting 11 seconds for this model to say its first word. It was painful. The best open model ever made and I kept closing the tab.
Not anymore.
This is the part of open source people forget. You release the weights and the entire industry fixes your problems for you. Same day.
Kimi K3 just became a real vibe coding option.
English
@jamiuxxvi MOTM is basically the only thing you need to see/know. MOTM in like 40% of games played lmao
English
Who even started that debate in the first place ? 😭
English
@blader I guess Fable 5.1 will drop very soon, reclaiming noticable lead over Opus 5? Otherwise, the model has no place. Also, Mythos class SHOULD have the highest post-training potential.
English
T. Rebentisch retweetledi
Benjamin Netanyahu is a war criminal.
Indonesia
@JacobPhillip801 @AlienWisconsin @DavidOndrej1 Obviously that will not happen in 2-3 years. How do you even think that would work?
English
@AlienWisconsin @DavidOndrej1 You are correct, as things currently stand. But chips are getting much much better at an accelerated rate with all the demand and research being invested. The point I’m making is who’s to say in 2-3 years we can’t have lower end hardware run 1-2T parameter models?
English
@bridgemindai @RodrigoVie52602 Yeah and now ask yourself why? They picked the only benchmarks where their model can compete. Its way smaller than Fable/Mythos.
English
@RodrigoVie52602 This is pretty much the only benchmark they released.
English
Fable 5 is finally back.
I burned through half my weekly limit before breakfast and I regret nothing.
But this chart should terrify Anthropic.
GPT 5.6 Sol Ultra: 91.9% on TerminalBench 2.1
Fable 5: 84.3%
Even GPT 5.6 Terra, the MID tier, ties Fable 5.
And GPT 5.6 drops any day now.
Anthropic's response?
Remove Fable 5 from subscriptions on July 7.
You do not win a model war by deleting your best model.
July is going to be the most insane month in AI history.
English
@ThePrimeagen Exactly 1/100 stanford maths grads can beat a calculator at any calculation. Being smart has no value anymore.
English
"being smart has no value anymore."
this is the mentality that paves the road to hell
MJ@mohitjandwani
@josssiiiah Brother exactly 1/10 stanford CS can beat Fable at anything. Being smart has no value anymore.
English
@AiBattle_ its official: the better model can easily be determined by which one can create the better minecraft temple!
English
GPT 5.6 will be released next Thursday on June 25
They have chosen the weaker GPT 5.6 checkpoint Kindle-Alpha instead of Kepler-Alpha
leo 🐾@synthwavedd
If you're wondering how people on your timeline seem to have access to GPT-5.6 Pro, it's now being stealth tested when 5.5 Pro is selected in ChatGPT (*at least for some Pro accounts)! I have also been told next Thursday is the planned launch date for 5.6
English
I found Opus
OpenAI@OpenAI
As AI takes on longer, higher-stakes tasks, we want models to carry beneficial and safe behavior into new domains beyond their training—and maintain it under pressure. That’s the idea behind our new research on training models to be broadly and persistently beneficial. alignment.openai.com/beneficial-rl/
English
@JesusGodAndKing @zkespresso @emanueledpt @elder_plinius you really think this? XD the "exploit" was just asking the model to fix a codebase without specifying if its security related or not. EVERY frontier model of the last 6 months can do that.
English
@zkespresso @emanueledpt @elder_plinius No this was found by Amazon Security Team and has nothing to do with Pliny, and it's not an issue that OpenAI has
English
So from what I understood:
The letter basically treats Fable 5 and Mythos 5 like weapons.
Every single use by a non-US person anywhere in the world needs individual government approval.
Cool.
So every new model we're gonna get is going to be like this?
Andrew Curran@AndrewCurran_
Bloomberg has a copy of the letter Commerce Secretary Howard Lutnick wrote to Dario Amodei on Friday. When the full PDF becomes available I will post it. Except: 'Until further notice, you must submit an application for an individually-validated license prior to the export, reexport, or transfer (in-country), including deemed export or deemed reexport, of the Mythos or Fable models to any destination worldwide or to any 'foreign person' wherever located'
English
@bridgemindai wym "for a fraction" XD its at least as pricy as GPT 5.5 which is 5 in 30 out. I mean yeah 1/2 is considered a fraction so "technically correct"
English
GPT 5.6 drops as soon as next week.
Insiders say it competes with Claude Fable 5 at a fraction of the price.
If true, this changes everything.
Fable 5 is the best model in the world right now and people are burning through $200 Max plans in 30 minutes to use it.
A cheaper equal would break Anthropic's pricing overnight.
Next week is going to be insane.
English