Angehefteter Tweet
Phunky
2.9K posts
Phunky
@phunkyflips
crypto curious, dipping into NFTs, DeFi, and other degen pastimes | profile pic @tubbycatsnft
Beigetreten Temmuz 2021
530 Folgt147 Follower
calling Nicholas Carlini "someone at anthropic" instead of "one of the best security researchers alive" is very funny to me
chiefofautism@chiefofautism
someone at ANTHROPIC just showed CLAUDE finding ZERO DAY vulnerabilities in a live conference demo claude has found zero day in Ghost, 50,000 stars on github, never had a critical security vulnerability in its entire, history... it found the blind SQL injection in 90 minutes, stole the admin api key, then did the exact, same thing to the linux kernel
English
Forever crashing out, this guy has never had a normal tweet
Under Secretary of War Emil Michael@USWREMichael
There are dozens of factual errors in the 42 page judgment rushed out in 48 hours DURING A TIME OF CONFLICT that seeks to upend the @POTUS role as Commander in Chief and disrupt @SecWar full ability to conduct military operations with the partners it chooses. A disgrace.
English
@flowersslop This is the closest Elon has ever felt to Icarus-ing himself. Karma can only be pushed so far
English
“AGI will make its own harness”
Yeah uhh… if that’s the case I think we probably get to a recursive self improvement loop before ARC-AGI 3 gets saturated
François Chollet@fchollet
AGI will make its own harness (or whatever else it needs to solve a new problem). As long as you need a human engineer to handcraft a task-specific harness/system for each new problem, AI isn't general. It's an automation tool to be wielded by software engineers. Harness-related research is important and valuable -- as a vector of better automation. But I don't think it gets us closer to general intelligence. General intelligence is when you can adapt on your own.
English
@AndrewCurran_ @tszzl @aidan_mclau I noticed the goblin/gremlin thing the first day and thought I was going crazy
English
@tszzl @aidan_mclau They don't know about goblin and gremlin.
English
i don't like or agree with this decision at all
that's not how agents work
Francois was like "tools don't exist, harnesses don't exist. WE'VE HIT A WALL"
like at least report the fricking scores with harness
Lisan al Gaib@scaling01
this is pretty much worst case performance no harness at all and very simplistic prompt
English
@kimmonismus Or… his statement is over two months old and the releases in that time were much more impressive for proprietary models than open source models
English
Two possibilities: Either he's saying this now as an employee of OpenAI out of loyalty (which I find hard to imagine), or option two, because he's seen how far along the models are internally (more likely).
Because his two statements strongly contradict each other.
Peter Steinberger 🦞@steipete
@sbaratelli @nvidia @openclaw most folks will want as much intelligence as possible, and open models aren't there yet.
English
OpenAI is deprecating the legacy deep research mode in ChatGPT on March 26 - the current deep research experience remains unchanged and historical conversations and results stay accessible
English
@Sauers_ This tweet blew my mind a little
Daniel McAuley@_dmca
some psychopath on the internal codex leaderboard hit 100B tokens in the last week
English
I am completely out of both Claude Max 20x and ChatGPT Pro rn 😭 good thing I still kept my normal Plus accounts too. I wonder how much Ant and OAI lab researchers and engineers use. I assume they have essentially unlimited usage
English
@flowersslop Grok you poor thing… the xAI RL environments might actually be the torment nexus
English
@scaling01 Aligns well with Dylan Patel’s statements on the recent Dwarkesh pod
English
massive models come at a cost
who do you think is iterating faster and RL-rollout-mogging all the others?
Grok-5: 6T
Gemini 3.1 Pro: ~5T
Opus 4.6: ~2-3T
GPT-5.4: ~1-2T
Lisan al Gaib@scaling01
I would genuinely love for this to happen but many people think that OpenAI and Anthropic are already in a positive feedback loop and as we have seen with Gemini 3 Pro: a ~5 trillion param reasoning model won't magically be AGI (or for that matter a 6T param Grok-5) my base case is that OpenAI and Anthropic will pull further ahead xAI has less compute, less researchers, less data (no Codex, no Claude Code) and does not have access to models that literally speed up research (behind ~6 months) Google on the other hand is still in the race, being only ~3 months behind. they have the most compute, researchers, an infinite money glitch and the data
English
I gave ChatGPT for Excel and Claude for Excel a try on a very hard Excel file: macro-economic data from 1,000 years of English history across over a hundred tabs.
I think both did a good job, and I did not spot errors (though I only did spot checks). However, Claude was harder to check because ChatGPT tended to stick within the Excel app, building formulas and manipulating the data in the way a person would. On the other hand, Claude used Python and often pasted material into Excel for display purposes only, making it harder to trace or edit. If that holds, I think it will generally make ChatGPT more useful for serious users if you want to audit the results.
Prompt: "help me understand the relationship between the mix of agricultural products in the UK, GDP, and population, along with hours worked. I want this over the total period, and you should illustrate interesting trends with graphs and statistical analysis
English
@scaling01 This is a good call, there are so many things to revisit from the last 18 months
English
just one more nap bro
Rohan Pandey@khoomeik
a few friends are trying polyphasic sleep so they can supervise their coding agents 24/7
English
Skills are among the most consequential new tools for AI, and Anthropic just released a very impressive nontechnical Cowork Skill that builds Skills, including doing interviews & providing benchmarks.
I think you still need to add the human touch, but this is a big leap forward
English
Let’s say this is mostly (80%) output tokens.
That’s almost $1.2m in token costs in a week using commercial 5.3-codex pricing
Daniel McAuley@_dmca
some psychopath on the internal codex leaderboard hit 100B tokens in the last week
English