Distillation is not an attack. It simply a way to train a model. Not doing it when you are behind is akin to snatching defeat from the jaws of victory.
Thinking about, if I had a lab, I'd be trying to get training data from a combination of the open web, piracy and all of my revivals. I wonder how labs are doing that.
Ah yes, because if a person agrees with anything a chinese company does it must be because they love Xi. Get real.
Munching of the top student in a class is clearly prohibited. Distillation is not cheating, it's learning from your competitor. Akin to a company purchasing their competitor thingamajig to see if you can improve their own product.
It is an attack at a sufficient level of sophisticated analysis. If you destroy the game theoretic first mover advantage, then you destroy the economic incentive to improve things.
Be that as it may, it would seem absurd if we start calling distillation out as antagonistic, but don't do the same for the SOTA models being trained on human-created data.
Given that model distillation has existed since the early days of the current AI boom, and no robust defense has been demonstrated, the available evidence does not support your theory.
Nothing superficial about 120K USD donations. I have personally seen nothing but good vibes coming from Bun to Zig and Andrew. Re-writing is not "throwing under the bus" you make it out to be.
I’m sure lots of rich assholes take comfort from this unfortunately common attitude.
This entire blog post exists due to someone clearly needing to vent, try to work backwards from that. If things were as peachy amicable as you guys imagine, this thread wouldn’t exist.
That someone needs to vent is not proof of anything. People get their panties twisted for the tiniest things. I would expect that if Andrew had a real valid reason than it would appear in this post. But it's really nothing but personal attacks with little substance.
The guy in charge of it has demonstrated that he'll cut people off from accessing it on a whim, though, so it's not really cheap and easy access for the entire globe. It's access for the selected people.
Other Internet providers at least have true boards of directors, shareholders with decision power, etc. One person doesn’t have the power to snap their fingers and make decisions based on how much ketamine they’re on at the time.
I understand that you want Russia to have access to it without interruption but until there is some sort of "International law" regarding those newer ways of providing Internet, politics will win.
This is quite insane to me. If I compare the output of LLMs for python vs statically typed languages it's really not a good choice to go the python route. It consistently produce relatively garbage code along actually good code. My experience has that the better static typing you have the better the code becomes.
LLMs have made me move away more from python rather than into it. I'm very surprised by this experiences of the author. The article is all over the place as well. Going basically all in on Python because it is apparently better than Haskell for LLM use and than agreeing with someone that says Rust is the best.
Exactly. The issue is them being automatically in favor, that’s a really extreme measure I as an eu citizen did not know about, even though I’d say I’m probably more informed then the average and was made to study eu history, structure and laws in school. That kind of stuff is really weird, even weirder is that such an emergency measure can be activated without countries being in active war or the like.
This is beyond retarded. It's not a loophole, it's too f stupid to count as a loophole. This interpretation is a violation of the law. They should jail every EU parliamentarian who claims that this is a valid interpretation.
Sufficiently advanced stupidity is indistinguishable from malice.
It completely perverts the entire system. With this approach, any law can be passed by a minority; just phrase as a negative.
I propose a new law: "The EU is hereby not not not abolished" ohh look, it failed to gain majority vote! EU is over! Everyone go home! Look at how clever I am finding loopholes!
>That kind of stuff is really weird, even weirder is that such an emergency measure can be activated without countries being in active war or the like.
What's also just as weird is how the EU can just straight up sanction, debank and make European citizens virtually homeless overnight, without even a trial day in court. Absolutely insane.
This is some NAZI/USSR shit, except you don't get executed, you just get deranked and made homeless and maybe die from that on the whims of some bureaucrat in another country.
We gave the EU superpowers to "protect us" but never asked ourselves what happens when they use those powers against us when people vocally don't agree with their agenda.
Jacques Baud (Swiss) and Xavier Moreau (French) citizens, got sanctioned by the EU[1] on the basis of "acting as mouthpieces for pro-Russian propaganda and spreading conspiracy theories". Sanctioning means asset freezes, debanking and travel bans, virtually a slow and painful sentence to homelessness and death since you won't be able to travel, have a bank account, a job, own or rent anything under sanctions.
I don't have a problem with them sanctioning people, I have a problem when it's not done by a judge but a snap decision done by bureaucrats without a public trial/hearing where the individuals get a chance to defend themselves for what they're accused of, especially when it's just for the act of speaking, even if we disagree with what they speak.
Because otherwise the EU is no different than a monarchic or totalitarian dictatorship who has people executed for speaking bad things against it, and there's nothing stopping them from doing the same to you, me and everyone else here for wrong speak against the crown's interests.
Nobody can be guilty of anything without a fair trial. Some bureaucrats power tripping acting as judge jury and executioners without trial is the definition of totalitarian tyranny.
Nobody, even law abiding citizens, wants to be under such an unaccountable legal system. Justifying it because it was used "only" against pro-Russians is insane. Today it's against them, tomorrow it will be against anyone who criticizes the EU leaders. That's how boiling the frog works. That's how fascists always did it.
>virtually a slow and painful sentence to homelessness and death
OK, do you have information as to how close to homelessness and death these two are? Based on my reading I'm unsure regarding Baud, but my impression is Moreau is probably not significantly affected.
sure, but people shouldn't lie about what the current consequence of things are either. I'm pretty sure I'm closer to death and homelessness than Xavier Moreau is.
on edit: also I think the whole death and homelessness worry expressed seems sort of weird, because I think the normal order would be homelessness and death. I may die soon, but in that context I will not actually be worrying about homelessness. Maybe that's just me though.
Xavier Moreau is not in the EU so a travel ban that prevents him from leaving the EU obviously does not affect him. The sanctions only affect the seized assets.
Meanwhile Jacques Baud is a Swiss Citizen in the EU and the sanctions ban him from leaving the EU, which makes the EU a de facto open air prison for him.
de facto a person can take a train into Switzerland from the EU (France, for example) with no border control. Or arrive by car. Once in Switzerland flights out are possible for a Swiss citizen. So I believe your statement is untrue.
Very reductionist comment- if you're an elected representative and you leave early to take a vacation knowing you'll be missing votes, you're not doing your job..
This. When I need to take my summer vacation, I need to request it to my manager in due time so the team can plan customer deliveries accordingly, and in my last day before leaving I need to do a handover of my open tasks to whoever will do the work in my absence. I can't just spontaneously decide one day that tomorrow I'm leaving for 2-4 weeks on vacation with no notice and no handover to my team.
What's stopping MEPs from having to do that? Do they have literally zero responsibilities and accountabilities? Because their job is pretty critical for our society an security, even if a trained monkey could do it in theory.
Oh, so it's also your fault when you plan, and get approved a vacation, and then in the middle of it (lets's say on week 2 out of 4) you're notified that in two days you must be in the office, or, according to you, you should get canned?
I don't know... I'm not European so I don't really care, but I feel like there are some jobs that have an existing overlap of _duty_. I was in the military, and PTO was viewed as a privilege, and sometimes leave was cancelled, but that makes sense because of the position. Other civilian jobs, like firefighters, police, maybe some medical practitioners, might have this same thing. Politicians I would say is definitely one of those positions, where you should actually be in "public service". Officials in a democracy are supposed to be elected not because we need people to fill vacant jobs, but because we need people to be on duty to make the hard decisions.
Basically, I don't think politicians should be held to the same standard as some SWE making note-taking apps.
This is a deeply American (and Puritan) view of work, and I can say that as an American who works in public service.
PTO is not a “privilege.” In fact, it is a documented right as part of the employment agreement your
company makes you, when you sign that document about the handbook. It should be a legal agreement, but somehow we’ve convinced people their purpose in life is to work for 50-60 years for 40+ hours a week and then have maybe 20 years to enjoy life before they die, happy to be of service to the people.
Public servants deserve MORE time off and MORE money because they literally are ON CALL most of the time. Taking a vacation should be MORE Normal and votes shouldn’t require people to be in person.
You build your government the way you build your country - you should show the utmost respect for those in public service by treating them right and respecting their time.
I didn't say it was a privilege. I said it was viewed as such when I was in the military.
I wasn't talking about work agreements with civil companies. You brought that in. My comment was in regards to public service only.
I don't disagree about public servants having more time off or more money. But I believe they should be on call. If that means cancelling vacation to vote, so be it. You don't want to have that life style? Don't run for public office.
FWIW, I've had skin in that game. I was stop lossed when I had 4 months left on my enlistment for a 15 month combat deployment. That's public service. Elected officials should be present and ready to serve.
Strange response... Since when is the idea of ‘duty’ supposed to be ‘deeply American’?
I’m not American but happy to agree with the OP - being an elected MP is not just a job and if they want to take leave whenever they want to and miss critical votes then I’m sure they would have no trouble getting a normal job instead!
Surely the parliamentary sitting schedule gives them plenty of time off already where they would not miss votes (I know it does in my country)
It is my limited understanding this is other way around - you plan your vacation, you get it approved, then on your vacation an emergency vote to reduce your salary is called and you automatically voted yes.
Sure you can. What unstoppable force is going to prevent you?
You might find out there are undesirable consequences if you make that choice, but that is only if the employer decides to bring undesirable consequences. MEP employers in particular are generally apathetic about having an employee on staff. Extremely so — to the point that they won't even take a minute out of multiple years to say "hi" to the person they hired, never mind give any direction to the employee.
It doesn't have to be that way, but when the employer doesn't care, that is the way it will be.
Ever worked with Europeans? This is how they treat work. In their culture taking vacation and actually switching off is part of the job. I'm not making a value judgement but their approach to work is very different to what some of us are used to.
So they should prorogue the entire Parliament and they all take a month off together. Why the fuck should some particular province or riding or whatever miss out on representation on an important issue just because their rep is out for a week on an island somewhere? Could they not have a proxy / surrogate??
taking vacation and switching off is actually the sane thing for a person to do. the problem is that there is no provision for your post being covered while on vacation, and it's a problem in the enterprise/organization.
imo that a representative of the people misses a vote should just not be acceptable under any circumstances.
that's of course if we had to take "democracy" as an actually meaningful representation of the people. this was a cheap political maneuver to get a bill passed. happens everywhere. they would have gotten the bill through by any other means anyway.
that would be in an ideal democracy, though, in reality european parlamentarians don't even draft laws, they are only able to sign off on laws drafted by shady appointees of an equally shady and largely unelected (by the people) comission, take it or leave it. that, photo ops and declarations is their whole job, it shouldn't be too complicated to have a fucking substitute when on vacation.
The idea behind _representative_ democracy is that _we_ take holidays and our representatives do the work of upholding _our_ democratic values. These MEPs took their holidays before the parliamentary session was complete.
If we're talking about _direct_ democracy then we have to do the work ourselves but because we don't want to do that (because we want to go on holydays), we give up our rights to those who _want_ to represent us.
That, of course, falls apart once those representing us start to only represent themselves.
> Plenary sessions of Parliament take place 12 times per year in Strasbourg and last four days
There are 48 days in sessions over a year. You're saying that it's too much to ask MEPs not to take holidays during those 48 days? Am I missing something here?
Do you those 48 days are the total number of working hours these MEPs have? Of course not. Something always needs to give when going on vacation. And usually missing a a plenary session is not weird or the end of the world. The problem here is that in this case the rules were abused.
Most of the work of an MP is not sitting in plenary sessions looking bored. The normally important bits are in-between - committee and office work. This specific event here is an example of a plenary session suddenly being made extremely important by a procedural trick.
For doing actual work? Getting in touch with constituents, listening to them, faction internal coordination, working on committee specific stuff, understanding proposals and motions of others, maybe filing their own... when taken seriously, this can be an intense job.
That's precisely what a regression test suite is for. There is a bug, you fix the bug, you add a regression test. So if the test suite is well maintained these real world production scars are reflected in the tests.
By only one metric really, which is market capitalisation. SpaceX, which is famously shielded from Elon by the next level of managers, also produces a useful and profitable product.
The big problem is that a person making a mistake can be taught to not make that mistake again. That's also not foolproof but at least it works a lot of the times. AI are unteachable, if you have given them a good prompt and they do something wrong 90% of the time you are shit out of luck.
That is to say I do agree that building reliable processes out of unreliable parts with feedback is the modus operandi. However AI cannot meaningfully handle feedback and learn. And that is a key unsolved problem.
> AI are unteachable, if you have given them a good prompt and they do something wrong 90% of the time you are shit out of luck.
If the Model makes repeated mistakes on the same subject matter, you can update your agent.md file, or you can add skills to deal with specific prompts, or you provide a better default harness.
The whole idea of coding agents is their harness makes a big difference vs a pure raw model.
> However AI cannot meaningfully handle feedback and learn
How do you think models are created? They are trained on feedback and learn.
Its not cheap but you can post train models. This is how custom models are mode, that deal with specific tasks more efficiently and accurately.
Example ... Composer? Its base Kimi v2.5 model that has been post-trained 2 weeks, to create Composer 2.5, what is a much better coding model.
Its literally trained to make less mistakes by feeding it correct data. Hell, a lot of the models you are using, are often the same base model, where v2.0 was the initial released model but the model keeps training, so when they release v2.1, its still the same model, but with more training time on feedback provided to v2.0.
LLM Models are not a cake you cook one time and they are done, and you start from zero again. If you have the money, and a powerful server setup, you can take a model like GLM 5.2 and post-train it, to reduce specific errors. Sure, you need a ton of money because its a large model.
But people have been doing this with 5M, 100M, 1B, 5B models for a long time already. To the point that some of the small models can do specific tasks, almost or better then some of the huge more general trained models.
> If the Model makes repeated mistakes on the same subject matter, you can update your agent.md file ...
That's all just prompting.
> How do you think models are created? They are trained on feedback and learn.
No one is post training models on a single mistake. At least I have not seen it. I also doubt it is effective. Post-training on a single failure will not meaningfully change the model. That even sidesteps the entire problem that you don't even have access to models if you use a provider like anthropic/openai
Teaching modifies the learner. Prompting doesn't modify the model. It provides additional context that influences a single inference. A person who has learned something can apply it years later without being reminded. An LLM generally cannot unless the knowledge is incorporated into the model itself or provided again.
I don't think it's generally thought of an advantage that LLMs are stateless. In deployment it's nice, it would be much better if the model could continually learn.
I'm not claiming LLMs are not useful, they most certainly are.
sleeping helps learning, but people can learn without sleep. You can teach something to a person and they can usually do it immediately in some capacity. If you prevent a person from sleeping they will have reduced capacity to remember things. But it will most certainly not be zero.
I don't know of any modern workflows that rely on "we'll tell the person not to do it again", though. There's a reason that companies have adopted blameless postmortems, because if your response to the DB going down is "It's fine, Kevin learns and next time he won't misuse the prod credentials", you are guaranteeing prod will go down again in the same way at some point.
Literally the entirety of the worlds infrastructure relies on that. In the past we had (literally) had nuclear war hinging on a single person just deciding that some data point is an artifact.
Every modern workflow implicitly relies on that. No infrastructure is fully robust. There's a senior DB person who has learned many things many times over who could bring down most of the US power grid.
> The big problem is that a person making a mistake can be taught to not make that mistake again. That's also not foolproof but at least it works a lot of the times. AI are unteachable, if you have given them a good prompt and they do something wrong 90% of the time you are shit out of luck.
I feel like this line of thinking is kind of an unfair comparison. I'm not saying LLMs are magical beings that can suddenly learn by themselves after getting something wrong, but your "person making mistake then being corrected" assumes you do tell the person about the mistake and tell them to avoid doing the same mistake in the future, but for the "LLM making mistake" example you then intentionally avoid letting the prompt being changed in response to the mistake, which would be the "then being corrected" part on the LLM side of the comparison.
Similarly, if you just let a person make a mistake and don't let them know about the mistake, they might keep making that same mistake over and over again.
If you update how you use the LLM as you discover what mistakes it does, just like you'd correct a person, then you can use an LLM and also the LLM can "be taught to not make that mistake again".
I'm not against the prompt being changed, the point I was making is that an LLM is prone to the exact same mistakes even if you change the prompt. A trivial example is the very basic character counting mistake, I just asked chatgpt:
> How many p are in strawperry?
> There are 0 “p”s in strawperry.
And I can trigger the same mistake with various words even when adjusting the prompt many times. So I cannot teach chatgpt to correctly count characters.
> the point I was making is that an LLM is prone to the exact same mistakes even if you change the prompt
That'd be strange, what specific models?
Your example of "changing the prompt" doesn't actually tell the model what to avoid though, you're just rephrasing the question? Again you're not doing an equal comparison for the human, because in that example I bet you'd tell the human what they messed up and how to do it better, why not include that in the prompt when you compare these two?
I have a clojure/babashka environment that I work in. My agents/claude files specifically explain this and how to work with it and what to use etc. Fable decided it didn't want to use bb/clj and ran python to do cli work in one project.
So I update the files in all projects to specify that python is strictly off limits/not to be used/never to run it/etc and created a preToolUse hook -- because I learned that just because an agents file says something doesn't mean it gets followed. Hence the hooks, fool me once ...
Later, I start work in another project with the hooks and updated files saying to use the established bb/cli tools and to absolutely never run/use python etc. What does it do on the first task? Tries to run python, gets hit with the preToolUse hook error and replies with something like -- oops my bad I tried to use python even though the instructions told me not too. Every once in awhile in new tasks I still see the hook error at some point but it doesn't apologize like it did before it just switches to proper tools after it can't run what it wants even though it's in the instructions and could have been avoided.
"AI are unteachable, if you have given them a good prompt and they do something wrong 90% of the time you are shit out of luck."
please take a look at the error(s) made in the prior run. what could've been done better? create or modify an existing skill to emphasize this, or suggest additional language in AGENTS.md.
It will return a bunch of relevant-sounding insight, modify skills and context files… Then do the same error again.
We’re not at the point where AI is capable of knowing what went wrong and self-aware enough to understand how it could reliably change its own behavior.
For months I’ve been trying to have the agents stop manually writing our auto-generated SQL migrations and run the command that generates them instead. SOTA models insist on occasionally getting it wrong.
Indeed. Any meaningful AGI/ASI will have to have a form of memory / continual learning. Sam Altman said last year that this will be the focus for GPT-6.
The whole "soul.md" stuff today is a poor approximation to that. But I wonder whether it will grow into it, like chain of thought prompting grew into reasoning models.
LLM's as a technology - currently - are stateless. The memory layer is controlled by the agent. I m surprised with the gpt-6 reference unless it has to do with vertical integration between the agent and the layer.
> However AI cannot meaningfully handle feedback and learn.
Well this is the central bet of AI coding isn't it? We, the humans-in-the-loop, get better at knowing ahead of time which patterns AI will handle better than others, all the while the models actually get better.
> The big problem is that a person making a mistake can be taught to not make that mistake again. That's also not foolproof but at least it works a lot of the times.
You can do the same to an LLM with the same outcome.
reply