Aurornis

> My solution was to create a weekly "team meeting" from 8am to 11am every Friday and invite all of them to it.

I've been part of these and the focus time is nice.

It doesn't address the root problem, though. If engineers are being bombarded with meetings, those meetings just get squished into the afternoon time slot.

The secondary problem was that some people started treating the blocked-out time as their only work time. Once that dedicated block came to an end, they mentally switched out of work mode and were not going to do any more work until the next morning. Before all of the "nobody can do more than 3 hours of good work per day" replies, this was a real problem because people also used this time for discussion things with each other on the team, actual team meetings, and various things other than heads-down work, leaving less than the allotted time for actually working. It was also a problem when something would come up at 2:30PM and the person responsible would defer it to next morning's work session instead of taking an hour to do the work. This might all sound like minor annoyances and it was only some people doing it, but the theme repeated itself so much that many of us soured on the idea over time.

The bigger problem was that it didn't address the real issue. Every situation is different and in my case the manager was part of the problem. His solution to everything and answer to every question was to get two people talking. He thought of himself as a switchboard, routing people to other people on our team.

Blocking out meeting time helped a little, but what we really needed was an active manager who was handling communications himself and only bringing the team in when needed. We also needed him to have our backs when we declined meetings or asked if they could be an e-mail.

The healthier managers I've had take a constant, active interest in keeping their teams' meeting workloads down and rooting out interruptions and distractions. They also empower their team to forward meeting requests to the manager to evaluate and ask that external requests at least cc them for visibility. Every time I've had a manager like that, I haven't felt the need for blocked out focus time.

Deepseek Flash plans were very cheap when they first launched. They raised prices https://finance.yahoo.com/technology/ai/articles/deepseek-ra...

Your link is about a different model.

The talk about subsidies is confusing because for OpenAI and Anthropic people usually include their salaries, training costs, and everything else. When the topic switches to open weight models we pretend the models appeared out of the ether at zero cost, and the only cost is running the servers.

> You still have the same expected value.

Yes, which makes the lottery comparison moot.

> "Job interviews" in tech used to be casual conversations of a few hours. There weren't even quizzes to see if you really could program

How far back are you referring to? I'm definitely not the oldest person around here, but most of my interviews since graduating college were more than casual conversations.

The exceptions were primarily where I was coming in via a referral or they were hiring me for something that was already proven and verifiable, like open source contributions.

The only time I see people blitzing interviews and stacking 5-10 different companies interview processes is when they’re unemployed urgently trying to change jobs. In those cases it makes sense to do whatever you need to do to stack interviews.

For normal job changing, people are more selective and interview maybe a couple times per year, spread out over time. In my experience it hasn’t been hard at all to get 3 hours here and there for interviews. Tech company culture is usually forgiving about people leaving for appointments and other things so this isn’t nearly as hard as it is for some of my friends in other careers where it’s PTO every time.

I’ve also not actually seen many cases where people schedule 10 interviews at once and actually get the full interview experience at every one of them. People get rejected at earlier stages of the interview process or they start pulling out if they realize the job isn’t a good fit. It would be very rare for someone to schedule 10 interviews and actually invest the maximum time into every single one of them.

So I understand your concern, but in my observation this pattern doesn’t really happen. People have been doing much more involved interviews (travel, flights, hotels) for decades.

That’s not how it works. Interviews are like a funnel. If you don’t pass the first interview they don’t schedule you for all of the other hours just to keep going. You get dropped from the pipeline.

When a company says they have 9 hours of interview time, that’s usually the maximum time if you make it through all the rounds. If you get that far, you’re highly likely to get an offer.

You would also be surprised at how many people will actually do 10-20 long interview rounds to try to maximize their chances for the best job.

There are almost two different bubbles of candidates: Those who refuse to do anything more than a couple quick calls, and those who will rearrange their life to spend days putting in work for each job they want. I think that’s fine and everyone should do what works for them, but if you think 9 hours of interviewing is going to turn candidates away from a company like Oxide you’d be very surprised to see the quantity and caliber of applicants who line up for jobs like this.

I think you also might need to adjust your expectations now that AI interview cheating is getting out of control. A lot of companies are moving back to on-site interviews. I’d happily do 3 x 3 hour video calls more than flying and staying at hotels and renting cars to get to an office in another city just to interview.

The reason companies have that many interview hours spread across multiple slots is that then have fallout over the funnel.

They don’t have every candidate scheduled for the entire time of all the rounds. If you don’t pass earlier rounds, they’re not going to waste time (yours and theirs) hauling you through more rounds of interviews.

If you make it to the end of all of those scheduled interviews, there’s probably a very good chance you’re getting an offer.

It’s interesting to me that 9 hours of interview time is considered excessive by some people. Over my career I’ve done interviews that required plane flights, hotels, and multiple separate days on site.

I’ve also done interviews that were brief 30-60 minute casual discussions followed by an offer during the hiring craziness that happened in a few years in the past. Anecdotally, coworkers at those companies were a completely random range from smart and kind down to the meanest, most clueless people I ever worked with. They had high turnover both because they were always firing bad hires (hire fast only works if you also fire fast) and people were burning out from the perpetual stream of bad hires being randomly injected into the company after they talked their way through the short interview.

Looking back, the interview processes that took the longest were certainly not perfect, but the companies that had them were much better about maintaining a coherent culture and high bar for the coworkers I would get. Every new hire felt like someone amazing who I was happy to be working with.

Everyone has different priorities, but looking back I think the relatively small amount of time I invested into long interview processes (relative to the 2000 hours I work every year) was time well spent.

> You'd probably be really annoyed, if your ticket is a winning ticket, you split the jackpot and don't even get a life-changing amount of money back for it. If any of theirs wins, you likewise get a modest amount, but you weren't bothered about losing £1

Your lottery comparison isn’t logical. I think you’re just repeating the same misunderstanding I was trying to dispel above.

If you want to force the analogy, you would have to imagine a lottery where your win and the investor’s wins are tied together. Either you both win or you both lose. But to make the analogy actually work, you’d have to imagine a lottery where their ticket purchase increase the winnings proportionally.

So in this bad forced lottery analogy, you would have a smaller number of the overall tickets, but the same odds of winning as before and the payout would be the same as before.

You’re making the common mistake of assuming that the company could get the same outcome without taking investment money. It’s common to imagine getting that your 0.1% of a sweet $1 billion IPO without ever getting diluted along the way, but getting a company to that IPO point requires a lot of cash. If a company is lucky enough to have all the cash flow they need to get there then they can do it all without dilution. Most companies need to raise some cash to get there, though.

So raising the money actually lowers your variance of getting a large payout, because the cash is what enables the growth to that liquidity event.

If they wanted to IPO they could do it at any moment and have a swarm of bankers and lawyers knocking down their door to work 16 hour days to get it done.

The only reason they’re not doing it is by choice, because they want to find an optimal time to maximize their returns.

There is no “we’re trying to IPO but can’t”. If they were actually trying, it would be in motion.

A confusing thing about fundraising and dilution is that the new shares don’t take away value from existing shareholders.

If each share is worth $1 at the valuation used in the raise, then an investor adding $100 million gets 100 million shares for it. The shares aren’t taken away from anyone, they're issued in exchange for the capital.

So ideally the dilution is neutral to the value of the equity. In practice this is highly variable because the valuations are fuzzy numbers used for the raise, but you get the idea.

If a company can get the same growth without raising, that would be better because the proportional ownership stays higher. However, the reason companies give equity in exchange for capital is that they need the cash for growth and can’t get it on better terms anywhere else.

Deepseek also came with heavily subsidized plans, at least at first.

This is why there are so many comments confused that you spent that much money on Deepseek. Those who got on one of the discounted plans could use very large numbers of tokens for trivial prices.

There is also a strange double standard for accounting for open models. People will look at OpenAI or Anthropic and say that we need to consider all of their training costs and employee compensation when thinking about the cost to serve their models, but when the models are released as open weights those costs are ignored. So in that way, the Deepseek models are heavily subsidized as well, with the possibility of them being served by a different company that paid nothing to develop them.

> Quality was ok, seems slightly above Luna quality perhaps?

I agree that it’s about in line with what you can get from Luna or Haiku, but I give the edge to Luna and Haiku when it comes to tasks that require world knowledge. It feels like their training sets were just cleaner.

Luna and Haiku are also close to free with a subscription plan, and they’re even very cheap at API rates.

So the amazing thing about Deepseek Flash is that you can almost kind of get that level of performance from an open weight model. It’s not as amazing when you start comparing it for how we really use smaller models from frontier labs on subscription plans.

I don't know why comments pointing out this simple fact are getting downvoted.

The oversimplified view that has been drilled into startup discussions for years has been that IPO is the singular goal for every startup and they need to get there as fast as possible, but that hasn't been true for a long while. There are high profile examples like Stripe with no intent to go public any time soon. Some public companies are even gradually doing share buybacks partially to remove their public exposure.

Being a public company kind of sucks in many ways. I'll admit my sample size is small, but every post-IPO CEO I've known has expressed some regrets about going public. It was a fascinating revelation to me after being raised on the idea that IPO is the ultimate victory goal of every startup.

> OpenAI has no long-term viability and they're desperately hoping for some new breakthrough to reinvent their business model before the VC money faucets turn off for good.

If a startup is riding a hype cycle and is one of two leaders in the global industry with unreal growth numbers, they can IPO whenever they want. The incentives lean toward doing an IPO before the hype runs out, not delaying it.

If they were worried about running out of VC money, going to the public (the P in IPO) would be the move.

Companies don't actually have to go public quickly or even at all, even though that's been drilled into us as the only goal of every investor-backed startup.

> Where I am it costs $20 to properly dispose of an old can of paint.

That’s a good way to discourage proper disposal.

Modern latex paints are accepted in normal household garbage in most cities if allowed to dry out first, but oil based paints should be collected at proper locations. It’s actually getting hard to find oil based paints for sale, but you want to encourage people to dispose of them properly. When someone discovers 10 old paints in the basement of an old house it shouldn’t cost $200 to dispose of them.

All of the counties I’ve lived in (USA) long enough to have material to dispose of have provided free hazardous waste collection sites. The only hassle was maybe dealing with the government employee who didn’t want to help you do the mandatory sign-in and the minimal paperwork some required.

My current city also has annual hazardous waste pickup days. They’ll advertise in your neighborhood with flyers explaining what they will collect and where to take it. Very convenient to drop it off locally.

Charging people for small scale disposal is a very bad idea.

The title is as misleading as they could make it while being technically correct.

The study found that it offered different products to users depending on context inferred from their other data. If you have a history of buying expensive clothes and luxury items, you're going to be offered more expensive clothes and more luxurious items.

The chatbots were not showing different prices for the same products.

Yes, the original quote was narrowly directed.

He was making a pitch for Steam in a different time where digital distribution was not a thing.

Now digital distribution is the baseline and availability and ease of use are in favor of not pirating, but his quote is taken out of context to justify piracy anyway.

> - pricing - for some kids some games are just too expensive, and they still want to play, and don't want to be left out;

Sorry, but that’s life. Saying something is too expensive but you still want it doesn’t change anything.

> - platform lock - this is more for emulation, where people don't want to buy a platform to play 1 or 2 IP;

How is this relevant? You can’t play a pirated PS5 game without buying a PS5.

This is just another variant of your argument above that people want things but don’t want to pay for them.

> but you can't assume they would be customers at the moment they get a copy.

You can’t assume that every pirate would not be a customer if they couldn’t pirate.

Which means the sales would be higher without piracy. Which means piracy hurts revenue. The rest is hand waving.

Personally I wouldn’t have a problem with something like this for retro gaming for titles no longer available.

I’m not sympathetic to this argument in the context of current generation gaming for titles where developers are trying to sell them and making them easily available digitally.

> "Piracy is a service problem" - Gabe Newell

This quote has been taken too far out of context from the original talk, which is worth viewing.

He was saying that piracy can be reduced by making the product easily available. When it was easier to download a pirated version from home than to drive to the store and back and unwrap a physical CD, piracy was easier.

That is not true today. The easiest path is to buy the game digitally through the service on the device and play it directly. Pirating it is more work and harder to do.

So quoting Gabe Newell in support is wrong. Saying it’s a service problem is also wrong when the piracy route takes a lot more work. The goalposts keep moving around (and I’m sure they’ll move more in the replies) but we have to eventually admit that a lot of pirates are pirating because they don’t want to pay for the game. It’s the most obvious fact on this topic but the comments are always full of mental gymnastics pretending the pirates are noble people doing the right thing.

> I don’t see your comment as a valid dismissal. Just one more way that individual outcomes in our society are increasingly K-shaped.

I was dismissing the claim that the product will fail because the world doesn’t like it.

You changed the subject to something else.

The program is limited to a few prescriptions. I searched their website and found this about their AI prescription platform:

> Only topical acne medications from a restricted list: tretinoin, adapalene, clindamycin with benzoyl peroxide, benzoyl peroxide, and azelaic acid, some combined with niacinamide or topical spironolactone. Each plan includes a morning and an evening treatment. The pilot does not include oral or systemic medications, hormonal therapies, or isotretinoin.

So it's a short list of topical creams. Several of them are already available over the counter. Benzoyl peroxide creams are on the shelf of your nearest grocery store. Some of them are available in higher strengths via prescription only and should be reserved for only severe or refractory cases after escalation. A lot of people go straight for the highest strength they can find on the shelf, so having a prescription gate for the stronger stuff can be justified.

Others on the list like tretinoin are currently prescription only. That one is a strong chemical that will burn your skin if you don't read the instructions about to apply and also makes you prone to sunburn. A lot of people don't read the instructions, they just apply the cream and see what happens. Having a provider, even if it's AI, do something to force them to understand the risks can be arguably justified. Tretinoin is also a risk to fetal development, so having a provider ask if female patients are on birth control is important. That said it is OTC in some other countries, so there's an argument that it should be OTC here as well.

The AI doctor is not allowed to prescribe anything really serious. It's a combination of prescription versions of OTC creams and a few topicals that benefit from some light provider instructions. Having a $5 option to get them with the serious cases shunted to actual doctors is not a bad tradeoff.

The Qwen models, especially when quantized, can be really bad about just trying things and seeing what works. If you’re not watching all the tool calls you may not see it, but it’s kind of scary to watch them just bump into wrong decisions and backtrack.

They also have a bad habit of accidentally building URLs that hit Alibaba infrastructure, likely because their training environment had them use those URLs. If you haven’t watched the outgoing network requests you might be very surprised at what your Qwen agents do sometimes.

> What AI boosters don't realize is that we're not in the optimistic days of the 1990s anymore.

Comments like this are in a different reality than most consumers. Most consumers don’t care about things like avoiding lock in to a platform. If the platform solves their problem then they don’t have any reason to leave it anyway. They’re not worried if their data is used to show them more targeted ads.

Topics like this show a sharp divergence between what you read on Hacker News and how actual users operate. We already knew that people who don’t trust Facebook aren’t going to suddenly start using Muse. They were never going to consider it. For the people who do actually use Meta products, which is a customer base counted in the billions, many will not have any problem using these tools.

Notice the word “immediately” in that statement.

The journalists were in a rush to publish breaking news. They weren’t going to wait for a response.

This is just a CYA statement to say that they sent a message to the company to do their job but, quite literally, did not immediately get a response.

Anthropic could put a big flashing warning text at the top of every chat that says “We are spying on you and will report anything scary to the police!” and it would not make any difference in this case.

You can call it anything you like, but only the legal definitions matter for the legal case.

> Presumably, Anthropic did the spying and the reporting.

You don’t need to presume. Anthropic reported it.

“Spying” as a legal concept has a definition that does not apply here. You could say they were “spying” in the sense that they read someone’s input, but that’s literally what they said they were going to do in the agreement when the person signed up.

So I responded to the question about the case being thrown out for “spying” by trying to show that the word doesn’t apply in the legal sense. If you sign up for a service that says “Hey we’re going to monitor your chats and might report things to the authorities” and then they monitor your chats and report things to the authorities, you should not expect the case to be thrown out for “spying”.

I was responding to a question about the legal case. The police did not perform any spying.

Frustrations about Anthropic’s EULA are a separate matter.

> when you only read the threat by spying on them

The AI companies have clauses in their user agreements saying they can review content flagged as harmful. It’s not legally spying.

If you recall previous outrage about ChatGPT being used in cases of suicides or shootings, this is the result. Every time a crime was committed and the police found ChatGPT history about the crime, the media turned it into a frenzy. So the AI labs added safety filters to their consumer plans that detect threats of violence, escalate them to human review, and report to the police.

Spying is not the right analogy because the information was given to the police by a third party which had a EULA saying they would do this. A more analogous situation would be someone reading another person’s diary and then turning it into the police department. There might be some limitation in the law that makes the evidence inadmissible because it was not intended to be shared with anyone, but that’s a separate decision.

Doing more hard problems and learning how to persevere through them, even if it ends in failure.

There are some smaller strategies you can employ like doing introspection about decision processes that led you astray and learning how to avoid them. Reflect on what worked and what did not afterward. Try to recognize trigger points for both good and bad decisions then try to nudge your thinking toward or away from those decision points in the future.

The majority of it is simply doing more and persevering.