obblekk

There should definitely be a plan for the brine.

I'm optimistic we could:

* Build ocean pipelines that disperse brine over much larger volumes of water to minimize impact on wildlife

* Dump the brine water in engineered desert fields where it will evaporate and deposit salt reserves.

In general, we should be aware of the whole lifecycle of a solution, and optimistically believe we can solve each problem with more work.

I’d rather live in a society where companies can challenge the system without threat of shutdown.

Imagine how the other side would use the power to make contract decisions based on political loyalty.

We also get things like companies being able to refuse to give your data to the police without a warrant (unreasonable search and seizure protection for people).

I wonder if Elon is distracting people from the issue: can you go into an office, openly announce to everyone your boss is a terrible human being who should be denounced, and reasonably expect to keep your job?

Workers should have the right to improve their working conditions, but not to hate on coworkers. Be they bosses or subordinates.

Regardless, the unconstitutional part is this decision is like an illegally created new law - it essentially carries the weight of law, can only be appealed to the people who made the law, and was never debated or voted by democratically elected reps.

> … offers convincing evidence that the guards in the experiment were coached to be cruel.

This was the point of the experiment. That simply coaching people in banal ways to be cruel caused them to be cruel.

Yes the study participants were just doing what they thought the scientists wanted to see… exactly as many humans do in many human organizations.

Perhaps the guards believed deep down there was no real danger to the prisoners. Exactly as many people assume about civil society and the social safety net.

How many people would do things that are quite evil or disastrous to another human’s life if instructed to do so by a company’s performance management system, at risk of losing their job?

Remember the context in the early 70s was the relatively recent discovery of the Nazi’s unimaginably cruel treatment of people, and the unknown question of what could cause humans to behave that way.

It turns out, just some authoritative instructions.

Worth reading the original PIs 2018 response to recent criticism in full. Well written. https://www.prisonexp.org/response

Now that it’s been operational for a while, I would love to see an analysis of the James Webb to determine if there’s diminishing knowledge returns to size.

I worry a bit the cost of something like this may be better spent on other space exploration efforts which will yield more knowledge.

For instance, voyager 3?

How long does the job get to run?

Would be really awesome if this could be used to offload long lived jobs from the request.

Defer.run is building this for vercel and it’s really cool.

Very cool, I've been thinking about the need for a service like this.

Does the 22% fee you're charging mean a 22% revenue share of what the customer pays? If so... that's very expensive.

This is an interesting example of a highly intelligent person actively convincing people to remove some alignment constraints.

Unlike China, the US has prioritized decentralized growth for centuries.

Throughout the Midwest, there are excellent research universities, large industrial manufacturers, and smart hardworking people.

It would be folly to try and relocate existing expertise into a concentrated area already struggling with crowding.

Serious question - I read that Sweden has a lot of social issues right now, but then I see this "bipartisan" reform (is this word used in non-US politics?) of an important sector in a reasonable way.

What is the Swedish political system doing right?

“Moreover, incapacity is a near certainty for all people: there will be a period however brief, in which a person can no longer manage their own affairs, requiring help.”

I’d never thought of it this way, but it’s a thought provoking reality.

It's possible to build an EV or even smart ICE that has a bunch of modern features and doesn't phone home constantly.

But I think we've now crossed a threshold where most of society agrees with continuous monitoring of some things for safety reasons. Cars are probably the most technologically advanced, potentially dangerous, and socially disruptive thing most people buy.

If cars were invented today the regulations would be far more onerous and intrusive than the regulatory system we got from before computers existed. Safety regulators in the US are already evaluating next-gen regulations to require a car to disable if alcoholic breath is detected, regardless of driver overrides.

For better or worse, we're on a one-way ratchet with this stuff.

These photos are beautiful. But none of them strike me as "photography" in the traditional sense of "objective capture of the light at a scene" but rather all look like hyperrealistic digital art. If it was labeled as Dalle3 output I'd believe it.

That's not at all a knock on the photographer. There's a ton of creativity and skill that goes into producing these, particularly at the high res in the gallery display.

But it's interesting that photography and digital art are converging to a similar point. We're not too far from a high res raw image + DalleX being able to produce any imagery this artist + iPhoneX could create.

Here's my example of a traditional photograph: https://upload.wikimedia.org/wikipedia/commons/thumb/7/7c/US...

For written text, the problem may be even harder. Identifying the human author of text is a field called "stylometry" but this result shows that some simple transformations reduce the success to random chance [1].

Similarly, I suspect watermarking LLM output is probably unworkable. The output of a smart model could be de-watermarked by fine tuning a dumb open source model on the initial output, and then regenerating the original output token by token, selecting alternate words whenever multiple completions have close probabilities and semantically equivalent. It would be a bit tedious to perfectly dial in, but I suspect it could be done.

And then ultimately, short text selections can have a lot of meaning with very little entropy to uniquely tag (e.g., covfefe).

[1] https://dl.acm.org/doi/abs/10.1145/2382448.2382450

Curious if Scott Aaronson solved this challenge...

This is important.

Not just that amzn algorithmically finds the optimal market clearing price.

They may be finding an optimal dumping price to reduce competition and then an optimal profit maximizing price to capture monopoly profits.

If true, this would be a clear violation of anti trust law, despite being algorithmically implemented at scale. Unclear if intentionality matters for the law - even if the algorithms learned this behavior implicitly as the solution to a legal objective, it could be illegal.

RNNs are the correct solution, but infeasibly expensive to run.

A different way to think about it is Transformer models are trying to predict which part of the RNN network is "worth" keeping given a resource constraint.

Transformers use a simple heuristic today (and this result makes the heuristic better). Just like many NP complete problems, there might be approximations that are not perfectly correct but still useful. Transformers prove that is the case for neural networks.

Knowing the people or knowing the rules?

This is fascinating because the effect size is so large and so discontinuous.

That the US government has become so detached from actual law enforcement makes me wonder if there's anyone thinking about reforms to the political system itself. I'm not talking about politically charged stuff like redistricting or ranked choice that just shifts the party in power.

I mean ways to create more citizen power over their representatives. What would today's equivalent to "direct election of senators" be?

I was a PM at instagram in 2016 when we got a lot of these complaints from celebrities and short usernames.

Some users were getting hundreds of reset emails/day triggered by random people in the world trying to reset their password.

It's a really hard problem to solve because if these users actually forgot their password someday, they would really want those emails. We ended up creating a snooze for 30 days button at the bottom of the email as an imperfect solution to balance short-term spam and long-term lockout (with an override if the device id requesting the reset had recently been logged-in to the account).

Idk if that still exists on IG but doubt it was ever ported to FB.

@sama

Jk though I intuitively agree, I think this article doesn't really prove its claim.

> Are shared computers without user separation a thing? If so, these shouldn’t be used to access web applications with sensitive information at all, no matter how short the session expiry time is.

I would argue shared computers are one of the primary threats deterred by this.

Additionally, for the attacker has access to device section, the author points out an attacker could still access email. But, email services tend to be more developed and often have the ability to expire all tokens (or logout all browsers). The exact place short session expiration helps is when the service does not provide a feature to log out all sessions remotely. That would be a better deterrent by far, but may be more complex to build than a default timeout.

It's not charity, it's capacity planning but at a lower scale than before.

There's always a set of junior eng with high growth potential who are being trained to become senior. They will continue be hired, albeit with less simple work than before, because most companies do run scenario planning like "what if X left who is our backup".

The juniors who are not expected to ever grow to that level would no longer be sought out for simpler tasks as those tasks will be automated away.

Net impact is fewer total engineers, but those who remain are at a higher level of average skill.

I don't think so. Junior engineers will learn much much faster than the past (think about how much more effective GPT4 is as a learning tool than the "rubber ducky method" or manpages or even stackoverflow).

And a part of their role will morph into prompting GPT4 (much like this senior engineer has started doing).

If GPTx ends up in the narrow area where it's universally smarter than junior engs but definitely not capable of being a senior eng, then junior engs will just shift to the little remaining work for senior engs, shadow them for months to years like an apprenticeship.

Of course in that case the total number of eng needed will also decrease (already only a small percent ever get good enough to be considered truly senior), so there will be selection bias toward more intelligent engineers who are a step above GPTx. If none are left, then the profession will be gone and there will be no problem.

I've experienced a very similar feeling.

To me it feels exactly like finding wikipedia in 2005, or getting an iphone + wikipanion in 2008. The frontiers of my mind have been unleashed. A real bicycle for the mind.

Here are some tactics I use to "turn off gpt":

1. It'll be there tomorrow. The great thing about their threaded model is you can easily find the convo and continue it tomorrow. Remind yourself of that consciously (or tape it to your monitor!)

2. You're not behind, you're ahead. 80% of Americans haven't tried chatgpt. 95% of the world maybe.

3. Don't worry about juniors. They'll still be hired because now they'll ramp up faster and produce better code, using the same tool you're using. Same thing that happened when stackoverflow became popular and junior devs stopped "reading the source code" or "reading man pages."

For all the limitations of GPT4, it truly is great at coding. Exciting times.

An interesting modification would be to have it reflect on its own output each day, and build up a list of advice for future attempts, fed in the next day.

That would give it some “learning” and I’d be curious if

1. Would it converge to a consistent shape at all? Or just bounce around random shapes day to day

2. Would it produce unicorns more often than 1/118 times?

The hardest part would be getting it to interpret its svg output without seeing it rendered. The multimodal model getting a rendered image would probably be much better, but maybe not!

Build the habit.

When google doesn't return a good result to a specific question, switch to scholar.google.com and start reading abstracts. Everything may seem like an opaque maze at first, but just keep reading and patterns start emerging quickly and become useful.

Build the habit.

When google doesn't return a good result to a specific question, switch to scholar.google.com and start reading abstracts. It'll seem like an opaque maze at first, but just keep reading and it'll start clearing up pretty quickly and become useful.

This is the most exciting science event to follow recently.

Just wait until the normal media picks up on it. A story made for TV.

The researchers definitely believe it's real - they're fighting over who gets credit (hence the double paper publication).

Only a few days left until replications should start pouring in...

The method to produce this material as described in the related paper [1] is fairly simple and could be done at home with a $200 home metal melting furnace from amazon and the precursors (which also seem to be fairly standard easy to obtain metals).

If this is real, I'd expect some smart people from hackaday / youtube to reproduce this within weeks if not days.

If this is real, it'll change society quickly and permanently for the better. There's obvious wins in energy transportation and even generation, but actually having a room temperature superconductor is likely to result in an explosion of engineering use cases. It will be like the discovery of lithium ion which slowly transformed the use of energy throughout society, but faster.

Hopefully it repros.

[1] https://arxiv.org/pdf/2307.12037.pdf page 3