It did one more weird thing where it seemed to lose the ability to count in order:
=====
You're right that hard per-action gating causes loops. So rather than block the model, make the harness own the
critical structure deterministically, leave a git trail, and use soft guidance:
- (2) Commit breadcrumbs: /new commits the scaffold immediately (before discovery), then commits discovery
progress. This gives the trail you want and a recovery path — a rogue rewrite becomes a visible git diff you can
revert, instead of an invisible clobber. This is better than gating for your concern.
- (4) Deterministic, verified seed: finalize re-asserts the kanban (writes TASK-001 directly, ignores any model
mangling), verifies it landed, reports honestly. The harness guarantees structure regardless of what the model
did.
- (3) Soft-constrain discovery: prompt says "the .fractal files exist — never recreate them or write them via
shell; read_file before edit_file; only edit intent.md/constraints.md." Guidance, not a hard gate.
- (1) Fix command routing: the discovery > loop recognizes cancel/exit/:q and refuses a re-entrant /new — a
routing fix, not a model gate. This alone would have prevented Run 2's disaster.
I'm not going to start coding until you steer — I've burned enough trust guessing. My recommendation is to do (1)
+ (2) first (they're the highest-leverage and lowest-risk: command routing stops the trigger, breadcrumbs give
traceability + recovery), then (4), then (3).
=====
After that it seemed to behave itself and then did a compaction and since then it seems to be working properly again. Very odd. (and disconcerting)
Now it's arguing with me about why the edits it made aren't being picked up:
=====
I'm not convinced. I didn't run it until you told me it was all fixed up. I tried running it with this path
instead (base) spike@Tokenator music % python ~/Developer/spikefu/spikefu.py and it worked. It also now works
with the old relative path approach.
You're right, and I owe you an honest correction — I've now given you two confident explanations (Zed, then
"timing window") and both were wrong, since you only ran it after I said it was done. The fact that clearing
__pycache__ is what unblocked it means you were right: it was pyc-related. Before I theorize a third time, let me
actually test the mechanism on your Python rather than assert:
=====
I was happily plodding away with it earlier when it threw this out in the middle of a response in Claude code:
---
So — what did you actually see before you hit Ctrl-C? That's the信号 I'm most curious about, and it tells us what to
---
That's the sort of behavior I'd expect from a one or two year old model quantized down to about 1 bit - right word, wrong language in a response. Google translate tells me that's Chinese for signal. I wonder what caused that to happen.
Also, the one for gb shows Northern Ireland. Great Britain doesn't include any of the Irish island. That would be the United Kingdom. There's a quick overview showing the differences here: http://resources.woodlands-junior.kent.sch.uk/customs/questi...
Nice work though. I can see a lot of ways these could be useful.
The article states in the very first paragraph: "Let me tell you a little about the truck driver you just flipped off because he was passing another truck, and you had to cancel the cruise control and slow down until he completed the pass and moved back over."
The point is that the truck driver isn't doing anything wrong. He's just trying to get his job done.
Where in the article does it say he put anyone in danger or imply that he was doing anything that might warrant the police being notified?
In most EU countries people can be placed in a psychiatric facility against their will if they are deemed to be a danger to themselves or others. Paranoid schizophrenics are probably the most well known group who are likely to end up involuntarily committed, but there are plenty of others.
Once you have been committed against your will, you have to convince either a psychiatrist or a medical tribunal that you are no longer a public health risk. You are not free to just move to another facility or country.
Not saying that's what happened in this case, but the wording of the OP title, and from reading the translated article, it seems quite possible.
Not necessarily. I just finished optimizing an AES implementation where some of the functions are called hundreds of thousands of times when decrypting a 300k payload. In cases like that, micro-optimization is not at all pointless unless you forget to benchmark the whole thing for each version and compare.
I like the idea of the site, but it should be opt out by default for anything you scraped from the spreadsheet. Once someone claims their password, they can decide what info to display if any.
Scraping user data from one site and putting it up on your own without their permission is not cool IMO.
I agree with the negative comments about the $100/hr+ subtitle. Apart from the implicit elitism, it isn't even true for a lot of the people in the original spreadsheet.
Also, as has been mentioned in other comments, the reclaim password functionality is broken, so I have no idea what I can or can't edit on my profile.
Some things may be said in private mailing lists, face to face conversation, phone calls, IM etc. but the actual standards, the objections and the decision making process is all public.
It seems that the whole HTML5 standards process was broken from the start and is likely to remain that way. The recent hoopla is just some public airing of the dirty laundry that has always been there.
FTA: "Bars are wonderful at segmenting by demographic. Match the bar you’re going to with the user population you want to target. Different bars will produce slightly different results but the variation is not huge."
The problem wasn't with performance in general, it was with hardware decoding of H.264 video. I don't believe anything has changed on that front. The core animation change is supposed to help with vector graphics rendering.
You're right that hard per-action gating causes loops. So rather than block the model, make the harness own the critical structure deterministically, leave a git trail, and use soft guidance:
- (2) Commit breadcrumbs: /new commits the scaffold immediately (before discovery), then commits discovery progress. This gives the trail you want and a recovery path — a rogue rewrite becomes a visible git diff you can revert, instead of an invisible clobber. This is better than gating for your concern. - (4) Deterministic, verified seed: finalize re-asserts the kanban (writes TASK-001 directly, ignores any model mangling), verifies it landed, reports honestly. The harness guarantees structure regardless of what the model did. - (3) Soft-constrain discovery: prompt says "the .fractal files exist — never recreate them or write them via shell; read_file before edit_file; only edit intent.md/constraints.md." Guidance, not a hard gate. - (1) Fix command routing: the discovery > loop recognizes cancel/exit/:q and refuses a re-entrant /new — a routing fix, not a model gate. This alone would have prevented Run 2's disaster.
I'm not going to start coding until you steer — I've burned enough trust guessing. My recommendation is to do (1) + (2) first (they're the highest-leverage and lowest-risk: command routing stops the trigger, breadcrumbs give traceability + recovery), then (4), then (3).
=====
After that it seemed to behave itself and then did a compaction and since then it seems to be working properly again. Very odd. (and disconcerting)