ozr

I simply don't like a speech-based interface, but I think it's unfair to compare STT tech from even a few years ago to something like Whisper. It's dramatically improved in both accuracy and speed.

Not much faster tbh, but it's a bit of virtue signaling you're often required to do with generative AI.

This is what I mean by the (imo) unfounded cynicism about SpaceX. Failed 3 or 4 times, kinda sorta got one. That's a bad record for a certain type of engineering; but not for SpaceX, who is intentionally trying to launch fast, fail fast, and iterate.

It took _20_ launches to get a successful Falcon 9 upright landing. By 2021, across 31 launches they had a single landing failure. Across all of 2022 and 2023 YTD, 137 launches, their success rate is 100%.

By comparison, the shuttle program had 2 failures across 135 launches.

SpaceX is achieving higher levels of reliability for way less cost than anyone else.

Failure is part of the process and the process is clearly working.

I don't understand why so many people are so cynical about Starship being able to fly reliably when SpaceX has repeatedly demonstrated that they can build rockets.

It feels like if a prototype Toyota had a wheel fall off and everyone immediately assumed Toyota was incompetent and incapable of making cars. It just makes no sense.

A supermajority of the wealthiest people on the planet came from middle or upper middle class backgrounds.

He didn't 'mess up'. This wasn't an accident. Wire fraud has an intent requirement. The government has to prove beyond a reasonable doubt that the defendent acted with the intent to defraud, specifically, the 'intent to deceive and cheat'.

Not punishing this is the same as allowing it.

It's prosecutorial discretion. Generally speaking, they want to charge things that are a combination of big and easy-to-prove.

There's nothing beyond convention that would stop them from _also_ charging for creating companies under fictious names, for adding a wire fraud count for every e-mail he sent as one of the people, for not including their real address on any marketing e-mails they sent, etc etc.

You and the parent are both correct, which is the problem. Spamhaus is notorious for inconsistent behavior, driven largely by the personal whims of individual employees.

The main element of the crime here was creating imaginary people to 'sign' the affidavits submitted to ARIN. Phony companies, phony people, etc.

Then the jail time is calculated based on the value of what was taken as a result of the fraud.

5 years seems fairly reasonable for a multi-million dollar scheme, involving multiple employees, explicitly designed to defraud.

This looks like a decent non-paywalled summary:

https://www.cidrap.umn.edu/antimicrobial-stewardship/artific...

> Way back in 2014 or whatever it was I was really questioning the decision Fog Creek made to spin out Trello and Stack Overflow. It seemed to only benefit investors and I wondered how the companies could diversify their income streams with such fixed-niche products.

I was at Fog Creek (FC) while some of these decisions were made. There are a few missing pieces here that may help it make more sense.

The model was for FC to act as an incubator and fund other projects on the backs of FogBugz and Kiln. I'm more familiar with Trello, but with SO I'm sure the idea was the same: when you take VC, you are committing to a very specific model: burn cash on getting as large as possible.

The goal with Fog Creek was to _not do that_. So, rather than having FC take funding, Trello was spun out and raised as it's own entity.

FC employees received equity (in lieu of profit sharing) in these products.

The last product to come out of FC was the result of a few competing teams working on different projects. Glitch (nee Hyperdev) was.. not a great idea, with not a great team on it.

A leadership change shortly after the focus on Glitch led to (more-or-less) the complete collapse and acquihire of it by Fastly.

> Even though Trello was acquired, I feel like Fog Creek the company and all of its products have suffered the worst fates possible from this decision.

I guess it depends on your measure. FC employees made money. I regret what FC was turned into at the end.

DJB has a historical record of being correct on crypto.

His 'tantrums' led to Bernstein v. United States, which established source code as speech.

There's been innumerable runtime errors, both tiny and world-altering, as a result.

Ubiquitous is a good guess.

The silliness has been dying down lately, fwiw. This feels like a bit of a time capsule from a few years ago.

And, of course, now they aren't accepting crypto anymore to reduce their environmental impact.

Too much authority being given to people that aren't doing things.

This isn't the least bit shaky to well-adjusted adults.

I use a a chrome extension for it. Otherwise when I hit their wall on another device or browser, I just skip it. Reddit seems fine with killing themselves off, I'm not going to any great length to mitigate their mistakes.

The AI safety cult has some true believers. It's still fundamentally a grift.

This is simply wrong. Information that did not exist in the model can be added to a model by finetuning, both full and PEFT. It has been repeatedly demonstrated in practice and in multiple papers.

Fwiw, unpublished testing on LLaMA-1 13B showed that it was able to learn a new word and it's meaning via PEFT with <50 examples. Finetuning can unquestionably add new data to a model.

Jeremy Howard has written a bit about how quickly LLMs can pick up new concepts as well:

https://www.fast.ai/posts/2023-09-04-learning-jumps/

Reminds me of when the RIAA and MPAA were able to end piracy. Dark days.

Modern models are increasingly trained on synthetic data.

Ok. Now convince everyone else of that.

I don't think so. I'm not suggesting the use of violence as a deterrent[0]. My point is that the human urge to enact violence in response to wrongdoing has to be satisfied. If it's not satisfied by the state, it'll lead to mob justice and vigilantism.

[0] I think in many contexts it is, but that's distinct from my earlier post.

> How does the (in democratic societies) generally accepted principle in dubio pro reo fit into this?

By creating a society that, on balance, trusts the state and it's institutions to enact justice (apply suffering, in context).

It will fail to be perfect, same as any other system. But it just needs to work well enough that a sufficient mass of citizens don't lose faith in it and try to enact their own forms of justice.

We can all point to cases where murderers got away with it, or someone who was ruled guilty received a sentence we think is wrong, but _overall_, we're ok with the state managing violence on our behalf.

> In terms of practical social functioning, rehabilitation of criminals is much more valuable than punishment, which is why most democracies at least aspire to this ideal.

There is little evidence that any rehabilitation schemes are particularly successful, with the gaps between the most- and least- effective not being particularly meaningful.

You prevent crimes by removing people that commit them from society.

The alternative is for individuals to inflict suffering on each other, which is just vigilantism and mob violence.

The point is to avoid exactly that. If we can (mostly) agree that the state will punish someone sufficiently and appropriately, there's no need for vendettas.

> That’s right, I think that inflicting suffering on vicious wrong-doers is morally praiseworthy even in the absence of any deterrent effect.

Aside from the morals, it's essentially for a functioning society. The intrinsic urge to harm someone that has harmed you can't be allowed to happen. The state has to have a monopoly on violence to function.

If you are not allowed to inflict suffering on people that have wronged you, the state has to.

If the state doesn't, a large enough portion of people will do it themselves, and the state loses legitimacy.

Football, not soccer, of course.