rich_sasha

Project Innocence helps exonerate innocent people. They claim that about 30% of their “clients” had admitted guilt. My understanding is that it’s mostly based on DNA evidence so not really a matter of opinion: https://innocenceproject.org/exonerations-data/

“Why” seems to be essentially coercion from the interviewers. People told that if they admit guilt then they’ll get a slap on the wrist and be let go, else go to costly trial etc.

I do sometimes wonder what is sucked in when distilling open weights models. It is of course near impossible to know.

So suppose I have lots of money and some hidden agenda. If I train my network to be mostly very good at everything but subtly bad or even malicious at other things… then OAI or Anthropic or whoever else distil the model to accelerate their training… what then? Can I covertly smuggle something in?

At this point I’d pay money for a good course on how to use AI for coding.

I’m not a Luddite, and the thing I most need at work now is a mid-seniority dev to support me. I don’t care if their name is Tanner, Prateek, Claude or Luna, I’ll take the help, except the meat-based variant is, IME, near impossible to hire (beats me - so many devs in the market, but the CV stack is deeply underwhelming, and the interviews only worse).

But every time I try to use AI agents - and frankly I only use the frontier ones as work pays for tokens - it’s basically what the author describes. Lots of random changes. The model happily saying it is doing X while also doing !X and also Y and Z, and sometimes gets stuck in some weird loop. Last time (admittedly not a frontier model but not far off) it wrote a Python script which would write another text file, with the contents defined as a concatenation of chr(XX) characters. That was enough.

Fixing or reviewing all this nonsense, or babysitting the model line by line takes more time than doing the damn thing myself.

What makes it worse is the cadence of babysitting an LLM is really awkward. Again, I don’t know what Im doing wrong. But reviewing things every 10 mins destroys my attention span. Even a junior dev is better in this regard - give them something to work on, describe what you need, review 1-3x / day for a week, and you hope to get somewhere. Even if it takes a long time, it’s not a major effort, so long as you’re ok with the latency. LLMs are the opposite - low latency, high noise. Lots of little efforts frequently, and all requiring full attention - because it’s easy to miss the little “rm -rf /” or other Easter egg Claude left.

So in short, it matches the behavior TFA describes. And yet lots of people see to be able to derive a lot of value out of it.

I have a lot of success writing medium length snippets with it, in a conversational style. Sort of up to 500-1000 LOC in things that aren’t challenging. Or building a scaffolding for a webpage. Or a non judgmental, zero lag replacement to SE/SO. But a mid senior dev ain’t that.

It’s not just a rant, any suggestions welcome.

Successful people work hard and are lucky. More successful people work hard and are more lucky. The most successful people work hard and are the most lucky.

The hard work bit helps for sure but ultimately it’s optional.

This one is allegedly/hopefully a bug but just waiting for it to become a revenue stream / paywall. Uh oh, you want to run locally your dangerous illegal foreign open models, not licensed by Nvidia? Can’t work with you paying for a special security feature unlock on your GPU.

Or maybe just not at all - shell out for OAI tokens and send us all your data - your privacy is very valuable to us and we will make sure to be the only ones to monetize it.

You could say, Nvidia is selling pickaxes and has no business gatekeeping what rocks you hit with it, but I guess it depends on their relationships with major mining companies, and ultimately what pays more.

Huh. I’m just confused about the people who saw Trump v1 and voted him in again for v2.

> Damn man, cutting off my leg seemed like a good idea at the time, but it turned out bad. But a while later, I wasn’t happy with my life, so I thought, what the hell, I’ll cut off the other one too, see what happens. That was bad too! So when my neighbor asked me if I want to burn my prosthetic legs as well, I said - NEVER!

To be honest I think the experiment would be insanely impressive even if you prompted explicitly that the game is one of 4 options.

I suppose it’s almost exactly the right metric if you want to look at affordability of houses. Almost since a bigger household needs a bigger house/flat but only up to a point, a young couple needs barely more space than a young single person occupying their own flat.

But the company can just turn around and say, soz we don’t have your deposit. We invested it in Labubu dolls and they crashed in values. Also we lost them. Next time, DYOR.

If a bank collapses, you get your money back from the Fed, no questions asked.

So USDT is a great amazing decentralized censorship-free drop in replacement to USD except:

- much more awkward to use

- more liable to hacks

- deposits are not insured

- fewer legal protections

- doesn’t pay interest (or not easily anyway)

- and as I find out, bannable by a central institution with lower legal oversight than the Fed

I’ll keep a mental note of this.

How do you prevent it though? My understanding is that it’s genuinely hard in crypto. Any automated access control must be baked into smart contracts, which are very limited in terms of the logic they can implement (in practice - in theory Solidity is Turing complete Id imagine).

A bank API can be 10-100s+ kloc if they need that to implement access controls.

Yes that’s what I was referring to (I couldn’t remember his name).

Still my concern would be that the other Zuck would show he has no income and no assets, while a doctor or coder would end up paying extra tax on their retirement savings.

A nice idea I heard is to introduce a minimum tax rate based on wealth. Let’s say the magic number is 0.5%. Then you say, every individual (over some threshold etc) needs to pay max(normal tax calculation, 0.5% of wealth) .

What I like about it is that it doesn’t penalize people who already pay taxes. It’s a lower bound, not an extra.

What makes me weary about any such tax is that in practice, it seems to target the broad middle class - people who have enough assets to be interesting to the tax man, but who don’t have enough to hide it or avoid taxes.

Good point.

Maybe also then, importing module as Cython?

  import django with Cython

compiles on first run?

Not really. Numba is great at optimizing numpy code but not necessarily other general Python code.

Here’s an idea instead (maybe it already exists). Allow a decorator for writing inline Cython. It solves, I think, 90% of the problem. On first run, or eagerly, compile the decorated functions. Basically, Cython is the JIT, and user enables it explicitly.

This way I can easily write inlined functions with proper C speed, in a Python esque language.

Reasons why I rarely use Cython without this:

- unmodified Python code running in Cython is often faster, but not that much. 50% or so is my rule of thumb.

- but the development style is very different for Cython and Python, adding a lot of impedance. Eg reloading doesn’t quite work, profiling is different, you need a build directory and a build step at deployment etc.

- Cython proper is different enough and I’d argue unintuitive enough that if I have to have all these things in place, I might as well wrap my function in Rust.

Such a decorator would give you a low friction way to make things faster when you need it, with scope for making it much faster via all the funny Cython, while keeping Python-esque semantics, in particular "development semantics".

I think it would require some language modification since Cython was definitely a superset of Python; I think some of that syntax has been streamlined, but maybe still a strict superset.

Money is a medium of exchange. If I want something, I can get money in some way, then swap it for something else.

But if I have a near magical genie who can make me anything - software, farming business, physical things - and at a fraction of economic input compared to the legacy monetary policy - then the money is worthless. As the Aztec shells were to the conquistadors.

That they can’t print USD is irrelevant.

One thing that bothers me is that if the AI maximalism turns out correct (which is needed to justify the valuations), why would AI companies respect money and ownership?

So far, even major disruptions were still affecting relatively small segments of the economy. Car lobby in Germany, tech in the US, "Big Pharma/Food/etc”. They needed the rest of the economy.

But if some AI company figures out a way to monopolize everything: new companies, old companies, physical manufacturing - all the things they say they will, please buy here for lots of money - why would they be bound to pay out dividends or otherwise share the fruits of their economic labour?

I’m not saying specifically Anthropic would be looking to screw its investors. Rather, if you control an almost omnipotent genie, wouldn’t the temptation just be too much?

Or temptation aside - will money even have any value? If some AI company rebuilds the economy from scratch, and control everything from accounting to shoe making, why would they respect the value of some fiat tokens? It’d be like the conquistadors respecting the shells and beads of their new Aztec subjects.

Of course, perhaps none of this happens, but then the question is, how good an investment all of this even is.

BP share price recovered though so ¯\_(ツ)_/¯/s

Time after time I feel like this is the Luddite path.

Law enforcement can already do incredibly privacy-invading things, and we’re ok with it. House searches (with a warrant), arrests (with probable cause), etc. It’s tightly controlled and can’t be used on a whim.

I feel technology should absolutely be a part of policing, and also tightly controlled, just as these other powers are.

Do we actually know the bans aren’t working? TFA quotes a study in Australia based on self reported use.

In Poland, every now and then they publish self reported results on teenagers’ sex life. It usually turns out something along the lines of, 70% of 12 year olds have had 5+ partners, do not use any protection and took part in at least one orgy.

I can imagine the bans aren’t working but maybe asking teenagers "are you doing the cool banned thing" is not the right way to determine that.

So let’s double the speed - 1.2 Ma. You need half the mass now - 25kg/sec. If you want to have 10 seconds worth of hover - that’s 250kg of gas. Plus the weight of the storage tank. Plus - this is the amounts for hover. Not for any extra force on top of that.

It still doesn’t make much sense to me. And that’s before considering what happens when you get rocket stream hitting a tarmac road. Never mind public roads, this sounds like a major hazard anywhere. Plus the quantity of compressed gas son board (if that’s how it works)… not a pretty crash. At these numbers, imagine 10MJ released all of a sudden - apparently that’s about 2.5kg of TNT.

Surely I’m missing something… as I say, I can’t make sense of it.

The maths of the flying is a little puzzling to me.

Back of the fag pack, let’s say it is light for an electric car and weights 1000kg. I’ll also go on a limb and guess the air flow is actually subsonic - 200m/s would make it 0.6 Ma (still insanely fast!). Even then, that requires 50 kg of air every second. Or 40 m^3. Using everyone’s favorite metric, this empties an Olympic swimming pool in 1 minute. It also requires a MW of power. I mean… that’s a lot??? Though admittedly, only about as much as accelerating 0-60 in 1 sec (50% of that in fact). So the flying would draw as much power as accelerating at full throttle.

Power requirements would go down a lot with slower exhaust speed, but then you need to shift more fluid. 40 m^3 already seems like a lot to me.

I get it’s a neat little demo and they have to get SpaceX into the picture somehow, but still I don’t understand how it’s supposed to add up. Also, what do I know - just a humble physicist.

Amazon Prime is quite good, tbh. There’s a whole layer of complexity your life loses when you know you can get life essentials within a few hours with no planning.

As a cyclist, I tend to pay more attention to the bike shape than most people. Everyone knows what a bike looks like, but as with LLMs, most people would struggle to draw one - correctly connecting all the different tubes at correct angles etc.

I note that even these bikes are not actually rideable. For a bike to be stable, the front wheel axle needs to extend forward a little ahead of the fork - the fork must ”bend forwards”. This is what actually keeps the bike upright, not gyroscopic effects or some such. Given LLMs are generally good at basic maths, im surprised they get this wrong.

Maybe a bit less so in this case though. As per the announcements, the model was actually trained for this prompt, so was fed some training data. I can well imagine the data there was just not so good.

There’s a whole massive industry out there of companies looking to make 2% efficiency savings on everything, and VCs looking for this. The glossy presentation might have even said 2FA somewhere so yes it’s definitely safe.

The perspective of resiliency and how much will you lose if things go wrong seems kinda… new?

Funny that it should happen in France which has been at the forefront of European self resilience.

I thought the ultimate root of this situation is still 3rd Reich and the realization that, for the most part, the Nazis didn’t break any laws. They passed laws, legally, making all their horrid designs legal, then carried them out.

Which is a long way of saying, there’s a consensus that there are standards higher than national law we may want to uphold.

I know ICC is later, and also US never joined, in a somewhat hypocritical compromise. But at least the purpose and motivation, IMO, are clean and noble.

It would be in fact quite amusing if the technology the Neolithians were actually most proud of had nothing to do with stone, just didn’t last. Better fabrics? Waterproof clothing? Elaborate wooden tools made using “dumb” stone ones?

It’s a funny one. I’m not sure what a Lean-less LLM proof even is. LLMs are amazing at bullshitting and skipping key steps and details. I’d imagine a LLM non Lean proof to be generally hard to evaluate - harder than that of a human mathematician perhaps. And the scale effect is against OAI here - the firehose just keeps squeezing out proofs.

Well, fine - but my understanding is, if a fuzz-generated Lean proof is correct, that’s end of story. It can’t be “incorrect” if it “passes”.

You might think this is not very useful, maybe - but that’s not a reason to retract..?