Aashish Upadhyay retweeted
If Anthropic is merging more than 30,000 code changes per day, why aren't the GitHub issues being assigned and addressed?
This is a great question on r/ClaudeAI.
I think there's a bigger question:
Why aren't Anthropic triaging this with a software factor?
There's a risk that the issues could be used for prompt injection, but you'd have thought if any organisation has the expertise to build a software factory that can review the issues, filter out the risky ones and assign the ones that matter to either another agent or a human for further review fix, it would be Anthropic.
Building such a software factory and letting us see it work would make a better marketing case study than claiming to be merging 30,000 PRs per day!
EVMs were hacked. @elonmusk you may want to reach out to your Indian friend Rahul Gandhi, who’s is crusading against vote theft. And Ambani was also involved in influencing the US elections. Just so you know Ambani may be a boss in America as well.
It’s refreshing to view all videos here.
HackerTube hackertube.org/
Anthropic is offering companies up to $1 million of free token usage to run their models on their source code and find vulnerabilities.
Their tool can find up to 100k vulnerabilities. But it needs human validation to identify false positives, and fixes still need a human to check them.
There's a real opportunity here for engineers to use their human judgement.
They're trying to create a market for their product. Getting validation. Building a sales pitch to sell to other companies.
Aashish Upadhyay retweeted
Do you all realize what is happening?
Science is becoming privatized.
Theoretical science is increasingly becoming compute-driven, and similarly, experimental science will increasingly become robotics-driven.
Both increase the role of capital and reduce the role of talent. Ergo, fewer and fewer humans are going to be needed to do frontier science.
And when fewer humans are needed, global collaboration becomes less critical. The days of science as a public, academic good - with papers being transparently published on Arxiv and open journals might be coming to an end.
Now, in a globally open economy, this might have been alright, but fundamentally, this puts every country except US and China in a difficult spot. This includes India, UK, Japan, Korea etc. Geopolitical turmoil makes it even more obvious.
Notice how every single scientist that OpenAI is collaborating with for reviewing their latest AI-led science, is based in the US.
Will it be wise for Indian nuclear physicists to use ChatGPT models for their scientific analysis? All secrets are lost when you use cloud-based LLMs.
Some are starting to recognize this. (See @UKSovereignAI). I think we will see similar response from almost all of the top 10 countries.
Science, after all, is power.
Announcing three major scientific discoveries, one each in mathematics (harmonic analysis), computer science (complexity theory), and physics (quantum field theory).
AI starting to help us improve our understanding of the universe! openai.com/index/building-sc…
Aashish Upadhyay retweeted
We believe voice is the most natural way to interact with intelligence, so we created Dessa, the most authentic full-duplex speech model ever built. Talking with it feels like talking with a person…full of emotion, nuance, pauses, and a natural cadence.
block.xyz/dessa
When time is the scarce resource, burning other resources is the rational choice e.g. LLMs, F1 race.
But is time a scarce resource in Software Development? Or we are using software development as a first step to something bigger?
Tried copying Anthropic's playbook for scaling code reviews on my team. Opus 5.5 nailed the policy docs: a solid Review.md and Security.md. But, the performance was unimpressive in defining goals, metrics to track, and a practical feedback loop.
