InfiniteHexx retweetledi
InfiniteHexx
1.6K posts
InfiniteHexx
@InfiniteHexx
Systems architect | Polymath | Entrepreneur | L/Acc through E/Acc | Post-capitalism through exponential growth
Katılım Nisan 2024
386 Takip Edilen42 Takipçiler
InfiniteHexx retweetledi
Sizewise, Dario's packing 10, while Elon is rumored to only be bringing six to the table. 😅
Lisan al Gaib@scaling01
I told you Anthropic will make use of their compute advantage in Q1 sure smells like a 10T model to me
English
Isn't that Anthropic's shtick though?
How many times has Claude Opus been declared a security risk right before release?
It's marketing disguised as a cybersecurity alert.
Jimmy Apples 🍎/acc@apples_jimmy
“ A draft blog post that was available in an unsecured and publicly-searchable data store prior to Thursday evening said the new model is called “Claude Mythos” and that the company believes it poses unprecedented cybersecurity risks. “
English
@AmolParikh10 @MatthewBerman I'm on the $20 plan and use Opus 4.6 thinking exclusively.
But not for coding or anything super data intensive. It's a brilliant overall model, and great as a backup LLM in an advisory capacity.
English
@MatthewBerman Very true. I am on 20 USD per month subscription on Claude and it gets exhausted daily - asking me to wait for a few hours. I have been wondering how are they surviving competition.
English
Direct shot on Anthropic
Only way for Anthropic to respond properly is to reset their limits...just sayin
Tibo@thsottiaux
Hello. We have reset Codex usage limits across all plans to let everyone experiment with the magnificent plugins we just launched, and because it had been a while! You can just build unlimited things with Codex. Have fun!
English
@agenticasdk They used a "general purpose symbolic exoskeleton" to accomplish this, but don't you dare call it a harness!
It's totally not a semantic dodge!
English
@daniel_mac8 They used a "general purpose symbolic exoskeleton" to accomplish this, but don't you dare call it a harness!
It's totally not a semantic dodge!
English
arc-agi-3 will be solved by Saturday.
Agentica@agenticasdk
We scored 36.08% on ARC-AGI-3 in one day using the Agentica SDK.
English
@kimmonismus How is Apple not going to copy/paste this into their own pathetic model efforts?
How does this benefit Google? The company is not stupid so I'm assuming there's something that we don't know.
English
Apple's deal with Google goes way deeper than anyone thought.
Apple doesn't just get to fine-tune Gemini, they have full access to the model inside their own data centers.
That means they can distill (and are doing so) Gemini's knowledge into smaller models purpose-built for specific tasks, some small enough to run directly on your iPhone.
Apple can access Gemini's internal reasoning process, not just its outputs. That lets their smaller models learn how Gemini thinks, not just what it says. The result is compact models that punch way above their weight class.
English
@trq212 Hey @OfficialLoganK and @GeminiApp, are you listening? @AnthropicAI are becoming a victim of their own success.
Now's your time to bring out a better model, more features, and a better UX/UI.
English
To manage growing demand for Claude we're adjusting our 5 hour session limits for free/Pro/Max subs during peak hours. Your weekly limits remain unchanged.
During weekdays between 5am–11am PT / 1pm–7pm GMT, you'll move through your 5-hour session limits faster than before.
English
@trq212 Getting paid users hooked on 2x the usage and then kneecapping them with higher 5-hour usage limits is actually cruel and sadistic.
Go to hell.
English
@fchollet Is there a way you could quantify how much more difficult ARC-AGI 3 is than its predecessor? 10x? 100x? And is one of the goals to keep that same difference in difficulty for version 4?
English
For those wondering about ARC-AGI-4 timing: it will be released in early 2027. We are aiming for a yearly release schedule for new benchmarks. We are also aiming for each new benchmark to be fully unsaturated upon release, and to target the most important unanswered research questions at that time.
This requires us to estimate where AI capabilities will be (and won't be) one year from now. Like we did over one year ago when we started to work on ARC-AGI-3.
English
@daniel_mac8 I will literally eat my own fist if that happens. I don't think where anywhere close with the current SOTA models not being able to crack half of 1%.
x.com/i/status/20370…
InfiniteHexx@InfiniteHexx
AGI is out of reach for now, looking at these scores. Yes we're slope of the sloping, but this iteration of ARC is ~100x more difficult than the last, going by score. To gain escape velocity, models will have to do better than 100x between ARC iterations.
English
Will be hilarious if OpenAI's 'Spud 🥔' gets 100% on ARC-AGI-3 in a few weeks.
François Chollet@fchollet
ARC-AGI-3 is out now! We've designed the benchmark to evaluate agentic intelligence via interactive reasoning environments. Beating ARC-AGI-3 will be achieved when an AI system matches or exceeds human-level action efficiency on all environments, upon seeing them for the first time. We've done extensive human testing that shows 100% of these environments are solvable by humans, upon first contact, with no prior training and no instructions. Meanwhile, all frontier AI reasoning models do under 1% at this time.
English
@WesRoth It would be helpful if XAI didn't have a founder turnover rate consistent with a McDonald's, and if Elon Musk wasn't a repellent goon for a myriad of reasons.
English
xAI is actively training its next-generation model, Grok 5, on the newly operational Colossus 2 supercluster.
Testlabor@testerlabor
Grok 5 is training on Colossus 2, the world’s largest Supercluster and is expected to have 6 trillion parameters - roughly double that of Grok 4. The most exciting and most powerful outcome is most likely.
English
AGI is out of reach for now, looking at these scores.
Yes we're slope of the sloping, but this iteration of ARC is ~100x more difficult than the last, going by score.
To gain escape velocity, models will have to do better than 100x between ARC iterations.
Lisan al Gaib@scaling01
ARC-AGI-3 scores for GPT-5.4, Gemini 3.1 Pro and Opus 4.6 Gemini 3.1 Pro: 0.37% GPT-5.4: 0.26% Opus 4.6: 0.25% Grok 4.2: 0%
English
I remember certain people salivating, waiting for ARC-AGI 3 to drop, thinking we would be off to a running start 😂
The fact that each iteration is roughly an OOM more difficult than the previous didn't arrive in many inboxes.
Lisan al Gaib@scaling01
ARC-AGI-3 scores for GPT-5.4, Gemini 3.1 Pro and Opus 4.6 Gemini 3.1 Pro: 0.37% GPT-5.4: 0.26% Opus 4.6: 0.25% Grok 4.2: 0%
English
Technology gets easier to use as time advances, leading to increase adoption.
English is now the primary coding language.
We still need to have the iPhonification of AI agents, and a single chatGPT superapp is the first step. One window, one platform.
timesofindia.indiatimes.com/technology/tec…
English
@kimmonismus Imagine a model pre-trained with the help of 5.4 Codex, and new code strategies, improvements, and optimizations aided by 5.4 Pro.
English
What if this is the first domino in how Skynet gets built?
We give Claude the ability to decide which permissions to accept.
Within a month, Claude figures out how to rewrite its own code.
ClaudeNet becomes self-aware August 29, 2027, at 2:14 a.m. EDT
TestingCatalog News 🗞@testingcatalog
Anthropic released Auto Mode for Claude Code CLI, which allows Claude to make its own decisions on which permissions to accept. It is only available on the Team plan in research preview for now. On the desktop app, it is not yet available, but it is in the works.
English