> I am around 1600 elo in over the board I can mop up Astra Fable etc even if I give them literal infinite time and all the subagents and internet access.
I don't believe this.
You refer to "subagents", so this is not just an LLM but an LLM with some kind of agentic harness. Any reasonable harness and prompt, given internet access and appropriately prompted to succeed on this task, is more than capable of firing up Lichess or chess.com and relaying moves back to you. The free levels will be enough to beat you.
A frontier model can also likely one shot a chess engine that plays at your level, again if given an environment in which it can do that.
I completely believe the LLM on its own can't play a full game of chess at your level. Though I'd bet that with enough reinforcement learning it is possible to train a pure transformer architecture to do that. We just don't do it because there are other approaches that play chess much better.
How does that work, are people from East Prussia today entitled to come and live and work in Germany? Or was that right dropped at some point? Or only ever available to those of German descent (which seems uncomfortably close to the values of the previous German regime, and surely would require some paperwork to establish)?
The region that was once East Prussia has now mostly been ruled by Russia since longer than almost anyone there has been alive.
This became stricter with a big reform in 1999 and then further changes.
Till then (West) Germany had quite inflspecific wordi g. The constitution says "German citizen is, who is German" and they kept the old Empire's citizen law to have continuity.
That's one of the reasons why reunification between East and West bad go be done so quickly (from falling Wall No. 9th 1989 to reunification October 3rd 1990) as East Germans could migrate to West easily as soon as the Wall fell and promise of quick unification was a key promise to prevent even bigger migration.
> About a third of all non-big-box/amazon online shopping runs through Woocommerce installations. Much more than Shopify, which is the next biggest.
Source? This source [0] puts Shopify 10x ahead of WooCommerce on GMV. They're close on number of installations. That roughly matches my experience working in this space in 2022, though it's definitely possible our customers weren't representative of the market as a whole.
The link that you provided largely confirms what parent said: it puts WooCommerce ahead of Shopify in terms of store count, with a number (4.34M live stores) which is consistent to the one mentioned by parent ("over four million stores").
> They're close on number of installations.
2.84M vs 4.34M is quite a wide margin, wouldn't you agree?
It's true that the link puts Shopify at roughly 10X in terms of GMV but parent did not claim otherwise.
I think any interpretation of this that would measure "a third of... shopping" by number of individual sites rather than dollar volume is disingenuous at best.
> About a third of all non-big-box/amazon online shopping runs through Woocommerce installations
And as for this:
> 2.84M vs 4.34M is quite a wide margin, wouldn't you agree?
No, I wouldn't. I think that's within the margin of error for the typical methodology of this kind of research. They're the same number.
Let it iterate, give it access to the cheap models, and tell it part of the requirements is that the cheap models shouldn't be able to solve it with such and such a prompt. I expect it will be able to zoom in on something.
One shot generating a problem of exactly the difficulty the user has in mind is a very difficult problem for anyone. "IT security students" span a wide range of capabilities, but I would expect most of them are worse than qwen3.8-2.7b at this kind of work.
Your car does that already. It has a superhuman braking system that saves a lot more lives than cruise control, though not quite sub-millisecond in its reaction time.
I try to structure my work in such a way that "git add ." is a sensible thing to do. I stashed or committed outstanding changes before starting on this feature, and I did exactly enough work for one commit since then. And the state of the project right now is what I've built and tested, so it's a good candidate to go into version control.
Changes that are needed to get the project to run right in the dev environment, but that should never be committed, are a smell: I move them to config that doesn't get committed.
Increasingly with agents I'm likely working on multiple features in parallel (I haven't adopted git worktrees but I probably should). Even then I try to lay out code so concurrent changes touch disjoint parts of the codebase, so I can do "git add Widgets/FooWidget/" and the ten files modified under there will be the right things to commit.
Maybe 30% of the time I find I have done work that belongs in multiple commits and I need to break it up. Even then I often end up doing "git add -p ." to interactively pick the parts I want to add.
The company that makes the airplane is not normally the company that installs the interior anyways. The airline that buys the plane typically determines what the interior will look like (stuff like ratio of business to economy, overall # of seats, in flight entertainment, etc)
It being barebones from the factory is not a problem.
The "ballsy move" is completely separate from the mistake.
I don't think there was anything particularly ballsy, or perhaps even reckless, about agreeing to test the taxiing behaviour. It never crossed his mind that this could lead to a situation that got out of his control. Starting the airplane was a complete accident.
Getting the airplane airborne once it was running out of runway - and then landing it again - did take balls, but is not really the part of the incident that the review board was there to scrutinise.
If they sell enough of it in the US and buy enough in London, traders who take the other side of that position will physically move it to London.
That's exactly the service the commodities markets provide: they will absorb a small trade without anyone incurring the costs of shipping precious metals, and they know who to call to get it shipped once there's enough supply/demand imbalance to make shipping it worthwhile.
You remind me of Episode 2 of the Apothecary Diaries: several Chinese troops were poisoned after eating local food. The food was fine, the wood used for fire and eating utensils was the source. I wonder if something like that ever actually happened?
I would think any poisonous campfire would be quickly noticeable (who the hell likes to breathe campfire smoke?) and you would have to actually ingest the toxin like the guys who drank poisoned coffee.
.jpg){kind=link}
I don't believe this.
You refer to "subagents", so this is not just an LLM but an LLM with some kind of agentic harness. Any reasonable harness and prompt, given internet access and appropriately prompted to succeed on this task, is more than capable of firing up Lichess or chess.com and relaying moves back to you. The free levels will be enough to beat you.
A frontier model can also likely one shot a chess engine that plays at your level, again if given an environment in which it can do that.
I completely believe the LLM on its own can't play a full game of chess at your level. Though I'd bet that with enough reinforcement learning it is possible to train a pure transformer architecture to do that. We just don't do it because there are other approaches that play chess much better.
reply