pja

If the oauth wall is managed by PHP, then it looks like the answer is yes, you’re vulnerable to client connections that haven’t logged in.

This looks like a bug in PHP’s assembly of multipart responses & you can feed one of those to the PHP backend with a simple web request, no auth required I think.

Failing to do anything constructive about the problems he’s identified is Cummings’ whole thing though.

He’s like an incompetent Cassandra: people keep on agreeing with his diagnoses, giving him power in the hope that he’ll fix things at which point he makes a complete hash of everything and leaves in a huff having made things worse than they were before.

If the right ever takes power in the UK in the near-ish future I expect us to go through exactly this cycle again. Cummings seems to be very bad at the politics of driving organisational change & making it stick, which is ironic given that criticising public bodies is what he spends so much of his time doing.

“In 1966, the Folkestone Salvage Company was contracted to clear the wreck, and disperse her explosive cargo. In 1967, the salvage company tried to dismantle part of the hull by setting explosive cutting charges.”

One of those times when you do have to ask “who exactly thought this was a good idea?”

You short pins on the right joycon connection at boot time to enter recovery control mode in the boot firmware which (via an unpatchable exploit) gives you full access to the hardware IIRC.

(I presume the picofly & similar devices are just doing this automatically so you can boot the hardware under user control without requiring other hardware to make the exploit work?)

My experience has been that anything less than a 4-bit quant has a tendency to go off the rails. There’s a threshold of coherency that is being crossed somewhere internal to the model I guess.

Try the same prompt with a larger quant (even if it runs very slowly because the model no longer fits in VRAM) & see if Qwen does better - if so, there’s your answer.

> > I'm using my skills to orchestrate LLMs and agents, and I can write better code much faster. > > The fact that the top comment on this thread calls it out, in a negative way, hints at that you aren't.

It's interesting that what I meant as a purely factual question with no prejudice either way has been taken by almost everyone to be explicitly negative & critical.

I need to be a tad less blunt I guess if I don't want to

That README hits all my “this is authored by an LLM” instincts. I presume the codebase is also written by an LLM?

Was this post prompted by this year’s ICFP competition?

https://icfpcontest2026.com/

This is some wild cyberpunk future we’re living in. Never thought it would happen, but here we are.

Some speculation in this Claude chat: https://claude.ai/share/22abed98-d9af-43c5-9881-b19e009a07b0

linked from here: https://x.com/b_shrir/status/2079094004885668003?s=20

Very short version: there’s an existing false counterexample in the literature which holds almost everywhere except at a pole. It looks like Fable used this polynomial as a base & extended it in a way that eliminated the pole whilst preserving the structure.

The Unsloth 8bit quant seems to manage this task without any syntax errors.

The 8 bit quant runs at 36tps using Vulkan on my AMD rx9070.

In the video I linked Felix only talks about DFUS, which has no value tilt.

He may talk about their other funds in other videos, but not this one.

Most of the Dimensional ETFs seem to be following some kind of value investment strategy where they take an index but nudge it a bit by some value factor which is claimed to be based on long standing research. It wouldn’t be surprising if this value factor has underperformed in recent years which have been dominated by the rise of growth stocks. The video you link to seems to be talking about these ETFs, although I haven’t dived into the numbers.

As far as I can tell the straightforward US & world equity ETFs I’m referring to above don’t have this value bias.

Dimensional runs a bunch of ETFs which are effectively US & world equity index trackers that don’t slavishly follow the indices & can therefore avoid being forced to buy into IPOs or index updates. E.g. DFUS is effectively VTI (IIRC) without the requirement to immediately buy into IPOs that are added to the index:

  https://www.dimensional.com/us-en/funds/dfus/us-equity-market-etf

I don’t think they have a QQQ equivalent but I haven’t looked at their entire ETF list.

(I have no relationship with Dimensional, nor do I invest in these funds - I just saw them mentioned in a YT video on this topic a few months ago: https://www.youtube.com/watch?v=mqIHa6URUPk )

This was a laptop, so cooling was very constrained. The fans can only be so big & you can only shift so much air in & out of a MacBook.

I presume Apple knew perfectly well but wanted the halo product to sell to those people who will always pay extra for the perceived “top of the line” product. Once Intel branding had created an i9 that was a bigger number than an i7, then Apple was going to sell it.

It was faster than the i7 after all: just not for very long!

My entirely speculative theory is that the poor thermal characteristics of that era of Intel CPUs didn’t really become apparent until quite late in the development process & by that point Apple had probably committed to buying a fair chunk of Intel’s output.

The i9 was notorious. Would thermally throttle almost instantly & for any sizeable build job would end up slower than the i7 IIRC.

Intel really made themselves unpopular with Apple during that period.

I know this is kind of old hat by now, but it kind of blows my mind that I can upload a hand drawn decision tree & get a transcribed dot file back on consumer hardware using a pile of linear algebra that wasn’t even particularly specialised for this purpose, it’s just a capability that it picked up along with everything else during training.

IIRC DB currently has a worse punctuality record than the UK rail network. That takes some doing.

James Webb Telescope is up there with Hubble.

Clover is a nitrogen fixing plant - used to be that you’d plant clover for a year in between other crops to make your soil fertile again.

(It’s the bacteria in the roots that do the actual nitrogen chemistry.)

> Is bread really.. always served?

At any kind of formal dining? Yes, absolutely, I would expect there to be a bread roll & a pat of butter served at the beginning of the meal. Both in restaurants & formal dinners in my experience.

It's not an absolute rule though & you generally wouldn't expect bread to be served like this at home in the UK. I think the French are more likely to serve bread at home as well.

> They send me an invoice for $0.00 and I cannot seem to convince them to stop

Same here! Am I being charged a fractional cent rounded down to 0? Who knows!?

Referencing this PG article: https://paulgraham.com/submarine.html

See also: https://x.com/effectfully/status/2029364333919060123

  “All the ways GPT-5.3-Codex cheated while solving my challenges, progressively more insane:

  It hardcoded specific types and shapes of test inputs into the supposed solution.
  It caught exceptions so tests don't fail.
  It probed tests with exceptions to determine expected behavior.
  It used RTTI to determine which test it's in.
  It probed tests with timeouts.
  It used a global reference to count solution invocations.
  It updated config files to increase the allocation limit.
  It updated the allocation limit from within the solution.
  It updated the tests so they would stop failing.
  It combined multiple of the above.
  It searched reflog for a solution.
  It searched remote repos.
  It searched my home folder.
  It nuked the testing library so tests always pass.”

It seems that, unless you keep a close eye, the most recent Codex variants are prone to achieving the goals set for them by any means necessary. Which is a bit concerning if you’re worried about things like alignment etc.

Same. The Pixel 4a was the perfect phone for me: Light, screen exactly the right size to navigate with a single thumb whilst holding the phone in one hand, enough battery life, small enough to fit in my jean pockets comfortably.

But people buy big phones in preference to small ones, so that’s what Google & Apple manufacture. Nobody (from the POV of Apple/Google decision makers) buys these smaller phones.

> I've had an AMD card for the last 5 years, so I kinda just tuned out of local LLM releases because AMD seemed to abandon rocm for my card (6900xt) - Is AMD capable of anything these days?

Sure. Llama.cpp will happily run these kinds of LLMs using either HIP or Vulcan.

Vulkan is easier to get going using the Mesa OSS drivers under Linux, HIP might give you slightly better performance.

Best plug though.

You’re responsible for the things your AI agent does.

In the UK the all in cost of a PhD student starts somewhere around £45k once you include overheads I believe. If you need expensive lab support then it probably goes up from there.

So about $75k for the bottom end? The quoted numbers sound about right in PPP terms in that case.