Erik Meijer
@headinthebox
I don't think I am completely cognitively challenged, but there are two things that you all seem to master effortlessly that I cannot wrap my head around: - pitching, - and OAuth. Oh, also office politics.
Enough said ...
Prediction: a year from now, US computer science departments will find their enrollment at 50% of what they have been the past few years. I'm willing to bet a peppercorn on it for someone who really wants to hold me to it. (-:
Fantastic example of reward hacking in the real world. You cannot trust anything a model produces. Period. The only fail safe solution is to have the AI (legislative branch) produce an inspectable artifact and evidence that the artifact is safe/correct and have the proof checked…
When historical discrepancies pile up, models lose their way completely and come up with creative/fraudulent ways to balance the books. Instead of attempting to understand discrepancies, they start inventing fake transactions or pulling unrelated ones to pass the checks...
Kristin Cabot used to work at Neo4J. As Dijkstra once said: “The use of knowledge graphs cripples the mind.”
I’m winning :-)
We're starting to think about how to evolve OCaml to be a more effective language for agentic programming. One intuition here is that agents change what's important in 2 ways: - the toil of writing matters less - the ease of understanding and verifying matters more
github.com/chiphuyen/snif… is really cool. Attendees of the Claude Code event a while back all got 3 months free of Claude Ultimate. Turns out I used $512.79 on using Claude code to generate Kotlin bindings to the Claude API. Creating bindings for Llama was much cheaper at $178.12…
Monitoring must handle structure, not just tokens arxiv.org/pdf/2507.11473 Now obviously I am not part of the deep learning aristocracy, but I do find this paper rather weird and quixotic in many dimensions. The main flaw is that they propose to monitor the model's COT as is,…
Several people pointed me to the post about Kiro and “the future of software” and its relationship with proof carrying code in Universalis. While I admire Amazon's use of formal methods in general, calling tickets, UI sketches, or hallway chats "specs" feels like green-washing.…
Absolutely 100% agree. The worst thing is that once its IQ gets below a certain level, productivity sinks by an order of magnitude since now you are just cleaning up the shit.
Claude Code seems to have been a bit "dumber" the past few days and just making stupid mistakes. I legit would pay $500/mo to just have the good version that never gets a quantized/degraded during peak times!
* New Coke (1985) Tastes better than both old Coke and Pepsi. Tanked in the real world. * Wow Chips with Olestra (1998) Who does not want no fat and no calories. Tanked in the real world. * Crystal Pepsi (1992) Not to confuse with crystal meth. * Amazon Fire Phone (2014) 3D UI,…
I often joke that the only reason papers on C#'s and Visual Basic's language design got published at reputable conferences was because of kind folks like @GavinBierman and crusso.github.io from (then) MSR who added all the greek symbols and dirty pictures after the fact to…
Programming Language Theory has a public relations problem
The news says it is because of "Microsoft IP concerns". My guess is with Claude Code, Codex, Jules, ... coding agents are now officially a commodity. In which case it makes it makes sense for the founders to cash out by joining Google. And I cannot image that Zuck's hiring…
Big welcome to @_mohansolo and others from the Windsurf team joining Deepmind : )
news.ycombinator.com/item?id=445281… Mmm, happy my Lumina packet is still sitting unopened on my desk.