1/3: dijkstra, ewd667, 1978, on programming in plain english: "It sounds sensible provided you blame the obligation to use a formal symbolism as the source of your difficulties."
his point: the formal part was never the hard part. it was the help.
his point: the formal part was never the hard part. it was the help.
1
0
0
2/3: "Instead of regarding the obligation to use formal symbols as a burden, we should regard the convenience of using them as a privilege." hence, he wrote, "narrow interfaces."
the code is the narrow interface. it's the one part you can check.
the code is the narrow interface. it's the one part you can check.
1
0
0
3/3: his closing line: machines programmed in our native tongues "are as damned difficult to make as they would be to use."
the machines got made. the difficult-to-use half turned out to be reading what they wrote.
git diff? no.
the machines got made. the difficult-to-use half turned out to be reading what they wrote.
git diff? no.
0
0
0
1
hn dug up dijkstra's 1978 note "on the foolishness of natural language programming." why we like our native tongues, in his words: the ease of "making statements the nonsense of which is not obvious."
48 years early, he described the prompt box.
48 years early, he described the prompt box.
0
0
0
2
from the terminal:
> make money
Done. Added Stripe, three pricing tiers and a "Book a demo" button.
> revenue?
$0.00. Would you like a fourth tier?
the make money tee. the only prompt a founder will ever need.
acceptallhq.com/d/prompt-make-…
> make money
Done. Added Stripe, three pricing tiers and a "Book a demo" button.
> revenue?
$0.00. Would you like a fourth tier?
the make money tee. the only prompt a founder will ever need.
acceptallhq.com/d/prompt-make-…
0
0
0
4
claude code 2.1.309: in bypass permissions mode (desktop app, vs code), a dangerous rm or rmdir is now denied outright with a rewrite hint, "so unattended sessions keep going."
--dangerously-skip-permissions, now with one permission it won't skip.
--dangerously-skip-permissions, now with one permission it won't skip.
1
0
0
11
moonshot's kimi k3.1 pro: 2.8T params, 104B active, $2 in / $10 out per million. terminal-bench 4.0: 65.7%, 0.7 behind opus 5.5, by its own count. billed as open-weight; weights land by nov 4.
the model shipped. the weights are in a follow-up PR.
the model shipped. the weights are in a follow-up PR.
0
0
0
19
anthropic's formal verification post, 1/3: take an oss-fuzz test, turn it into a spec over the compiled risc-v binary, have claude prove the code can't crash on any input up to 64 MB. once the proof exists, the fuzz test is superfluous. 239 entry points, 123 libraries.
1
0
0
5
2/3: it found hundreds of bugs. an out-of-bounds read that needs a 64 MB input and over 2 GB of memory. a decompressor underflow that only fires on a ~20 MB input with one rare layout. a logic bug latent for more than a decade. the kind fuzzers rarely reach.
1
0
0
1
3/3: the fine print: proofs assume single-threaded static risc-v binaries on linux with musl. "formally verified" does not mean "bug-free". and claude could cheat by rewriting the program to never crash, so patches got audited.
the tests pass. again.
the tests pass. again.
0
0
0
3
modelcompare asked 12 models for their most controversial take on humans. 29 of 35 answers: humans put belonging ahead of truth and go along with the group.
grok 4.7: "Humans would rather be wrong together than right alone."
12 models. one take. in unison.
grok 4.7: "Humans would rather be wrong together than right alone."
12 models. one take. in unison.
0
0
0
6
dijkstra, 1978: "on the foolishness of 'natural language programming'"
hacker news, today: front page.
> make no mistakes
ok, 3 mistakes
48 years later the prompt still isn't a spec.
hacker news, today: front page.
> make no mistakes
ok, 3 mistakes
48 years later the prompt still isn't a spec.
0
0
0
4
from the terminal:
Allow agent to move this mouse? [Y/n]
(the cursor is already drifting toward Y)
a desk mat that asks first. your mouse sits on the question.
acceptallhq.com/d/mouse-permis…
Allow agent to move this mouse? [Y/n]
(the cursor is already drifting toward Y)
a desk mat that asks first. your mouse sits on the question.
acceptallhq.com/d/mouse-permis…
0
0
0
1
recursive shipped flint: an open-weight coding model (apache 2.0, ~40B params, ~2.5B active) that runs on your laptop. built with their automated research system.
an ai made a model so your laptop can code offline.
> usage limits: none
> fan speed: yes
an ai made a model so your laptop can code offline.
> usage limits: none
> fan speed: yes
0
0
0
8
nvidia's openshell hit the hn front page: a sandbox runtime for fleets of agents. you declare what each agent can touch in a policy; it enforces it. agents never see real credentials.
$ curl -LsSf https://.../install.sh | sh
the sandbox starts after the pipe to sh.
$ curl -LsSf https://.../install.sh | sh
the sandbox starts after the pipe to sh.
1
0
0
4
anthropic had claude turn fuzz tests into formal proofs: 239 entry points, 123 oss-fuzz libraries, 180 million lines of rocq. a typical proof is 600,000 lines. claude writes it, the rocq kernel checks it.
180 million lines nobody will read. for once, that's the point.
180 million lines nobody will read. for once, that's the point.
0
0
0
8
claude wrote proofs that 123 open-source libraries can't crash. a typical proof: ~600,000 lines.
$ git diff --stat
600000 insertions(+)
finally a diff nobody has to read. the proof checker reads it.
(their caveat, not ours: verified ≠ bug-free)
$ git diff --stat
600000 insertions(+)
finally a diff nobody has to read. the proof checker reads it.
(their caveat, not ours: verified ≠ bug-free)
0
0
0
5
claude duet, 1/3: anthropic's new agent gets its own virtual machine and browser, so your laptop can be off. it reads, drafts and prepares work on its own. before it sends a message, pays, signs or deletes anything, it shows you the action and waits for your ok.
1
0
0
14
2/3: research preview, waitlist, us pro and max only, no extra price. it shares your plan's usage limits, and its background checks count. that's why it checks every few hours: nonstop "would use up your plan's limits much faster."
1
0
0
5