I wonder if @emollick consulted the writers of the show Cupertino. It is surprisingly good.
36
How often does an internal tool compete with a spreadsheet?
Now, how often does an internal tool compete with a chat tool?
I’m going to make a bet that 70% of use cases for internal tools are glorified spreadsheets and ChatGPT/Claude with skills and mcp-ui can bring more value at a fraction of cost.
67
I don't think I ever heard the word 'canonical' being used as much as in the past year. AI is seeping in to our language.
28
Updated the University Challenge question bank with the repeat question detection:
maciek.codes/university-chal…
37
I created a question bank based on University Challenge episodes with some fun stats. Check it out:
maciek.codes/university-chal…
Fun fact:
@amolrajan is more likely to praise than Paxman
47
idk but it sounds super fishy and malicious.
eMudhra disabled FIPS on the HSM, exported the private key into Python on a computer, and then it got truncated when they re-imported it
I have to admit I did not realize CAs were running HSMs with the ability to do this. How many CAs are exporting private keys to computers??
55
Looking up what is on streaming services. Out of habit I googled it. And the page I opened quickly reminded me that I should have used chatgpt.
28
It's often the case in complex systems that it is not one decision that causes a catastrophic outcome, but rather a series of decisions that look benign in isolation.
sshhole.com/
1
30
ROI seems not bad given that Microsoft pays for LPE bugs (importantly and critical) $500-$40,000 in bounties.
We are now allocating a significant amount of our frontier compute towards discovering and patching vulnerabilities in open source codebases, fully pro bono.
In one case, we studied the Linux kernel, which is a particularly load-bearing piece of software at OpenAI.
Before AI, security researchers might discover ~20 local privilege escalation (LPE) vulnerabilities in Linux in a typical year.
In one week's time, our upcoming frontier cyber model was able to find ~1000 LPE vulnerabilities using about $1M of tokens. In the process, it was able to discover at least 10 novel classes of Linux exploitation techniques.
48
Today was my last day with Anthropic. I didn’t renew my subscription. I’ll give OpenAI a shot and then maybe come back soon. And work takes all my attention anyway. Remember, attention is all you need.
1
89


















