computomatic

I find myself wondering about this though. Because, yes, what you say is true. Transformer architecture isn’t likely to handle negations particularly well. And we saw this plain as day in early versions of ChatGPT, for example. But then all the big players pretty much “fixed” negations and I have no idea how. So is it still accurate to say that understanding the transformer architecture is particularly informative about modern capabilities?

Huh, and there I would have assumed this was defensive architecture akin to the Great Wall of China (albeit more rudimentary). I’m guessing that was ruled out early.

Have you tried nothing at all? Had great success with this on a 150+ dev team. Much preferred to jira. Admittedly does require a different approach to work than a jira-centric team is going to be familiar with.

> I though of pasting in Perplexity's summary, saying it was from Perplexity but that I had checked and it was a good summary.

> Would that be OK or would that count as an AI written comment?

The rule seems written to answer this directly.

Absolutely nobody cares what Perplexity has to say about the case - summary or otherwise. If you mention what the case is, I can ask claude myself if I’m interested.

Better yet, post a link to an authoritative source on the case (helpful but not required).

At minimum, verify your info via another source. The community deserves that much at least.

An AI-generated summary adds nothing positive and actually detracts from the conversation.

Not the parent but I infer “fresh” as meaning a new approach to an old problem (with the benefits of experience baked in). A synonym of “modern” without the baggage.

Sorry, we’re only considering applicants with a proven track record of failing as CEO

16 random bytes is not a valid UUIDv4. I don’t think it needs to be in the standard library, but implementing the spec yourself is also not the right choice for 99% of cases.

Seems to me the most obvious answer is to return to sundials, no?

I don't have a source to site, but I'm fairly certain the Canadian government is adopting (and presumably encouraging provinces to adopt) a general policy of explicitly not allowing US preferences to dictate our domestic policy moving forward. Of course, that is indeed in response to recent actions from the U.S. And in that light, this time change was an obvious early move as the only thing preventing it was the trigger based on the US states.

Cable at least made sense on paper (if not obvious to the consumer). The channels were independent companies, they pay for the rights to content and get paid by ads. But they had the problem of how to actually get their feed into your home (over the air broadcast was the only D2C option).

The cable provider was just a delivery mechanism. So you pay them to deliver the feeds. But they didn’t get any revenue from the content providers (or their ads).

In other words, two different companies, two different services (content vs delivery), and two different revenue models.

If I say “you are our domain expert for X, plan this task out in great detail” to a human engineer when delegating a task, 9 times out of 10 they will do a more thorough job. It’s not that this is voodoo that unlocks some secret part of their brain. It simply establishes my expectations and they act accordingly.

To the extent that LLMs mimic human behaviour, it shouldn’t be a surprise that setting clear expectations works there too.

From reading tea leaves, I’m fairly convinced Microsoft sees streaming games as the future of Xbox.

There’s a clear line from AI to that when you consider how many slightly outdated but not useless GPUs will be in Microsoft data centres within the next couple years.

That’s not how I read it. This describes a process of tagging, not summarizing. The tags (“clusters”) have a title and a summary, but those are not derived from the conversation. They are common across all conversations. Isn’t that what they are saying?

There might be some risk of some data leak where a new cluster (tag) is defined. But that’s not the same as saying they are viewing summaries of content.

On the other hand, species go extinct with increasing regularity.

Most, if not all countries have their own domestic payment systems. This is about cross-border payments within the EU.

“Breakup” seems a bit exaggerated considering the % of payment volume which might switch to the new system.

I think it’s more helpful to discuss this in requests per second.

I’d interpret “thousands of people hitting a single endpoint multiple times a day” as something like 10,000 people making ~5 requests per 24 hours. That’s 0.5 requests per second.

> merely matching the cost and harms of the existing system is far from enough

The new system needs to be better but that doesn’t necessarily mean safer.

For delivery, that could mean cheaper and faster and more convenient.

Autonomous vehicles are a special case because those accidents tend to cause death and serious injury. As long as delivery drones can avoid killing multiple people per year, they are probably fine to compete on other metrics.

This is great product design at its finest.

First of all, they never “handle more requests than they have hardware.” That’s impossible (at least as I’m reading it).

The vast majority of usage is via their web app (and free accounts, at that). The web app defaults to “auto” selecting a model. The algorithm for that selection is hidden information.

As load peaks, they can divert requests to different levels of hardware and less resource hungry models.

Only a very small minority of requests actually specify the model to use.

There are a hundred similar product design hacks they can use to mitigate load. But this seems like the easiest one to implement.

This matches my experience fairly accurately except for the one guy I met at a housewarming who handed out cards to everyone. It was so weird - I haven’t seen anyone do that in real life. He had a shop that repairs chipped windshields.

And you know what? About 8 months later my windshield got sprayed by gravel. That guy got the business (he’s a friend of a friend after all, and I had his number in my wallet).

I’d say the issue isn’t that cards are outdated. It’s that people aren’t using them correctly.

Presumably it corresponds to Table 8 from this doc: https://cdn.openai.com/pdf/8124a3ce-ab78-4f06-96eb-49ea29ffb...

If that’s the case, it’s mislabelled and should have read “17%” which would better the visual.

You don't think the LLM made the leap to California Academy of Sciences specifically because Coit Tower was in the context?

If you asked for driving directions from your local walmart to cal academy, the LLM seems just as likely to decide you mean something else by "cal academy" and use the tools available to determine what.

ETA: Out of curiosity, I tested this on ChatGPT. It did ultimately give directions to California Academy of Sciences. However, during research, it also checked the website of the local school district as well as Waze to determine if I meant other nearby schools, even looking up directions for a local acting school and daycare (each with "Academy" in the name).

This is a fantastic article with some facts I didn’t know. I hadn’t thought about how the self-closing tag is effectively nothing more than a stylistic choice.

That said, it failed to persuade me. I’d still favour the self-closing style in any style guide. At least now it would be a more informed (and better scoped) preference.

Considering two style guides:

   - <tag> is an opening tag
   - </tag> is a closing tag
   - <tag /> is a self-closing tag
   Note: it remains the responsibility of the author to know which tags are self-closing.

Seems more useful than:

   - <tag> is an opening tag or a self-closing tag
   - </tag> is a closing tag

Admittedly, it’s subjective. That’s why any consistent guide is better than none.

I did the same. The only way to win is not to play.

Regarding the inevitable degradation of open core, is your experience that they get actively worse or just stale? How would you balance that against something that’s good enough today for what you need today? (Ie, is being good enough today enough signal that it will stay good enough in the future if requirements don’t expand?)

The salient point from the parent comment was that “using” assumes the resource is an Object.

The core value prop of both “using” or “defer” is that a single scope can manage the lifetime of some resource while child scopes interact with that resource.

“Using” assumes that resource is a local object but that’s rarely actually the case - as in the example of a DB connection - it’s just how OOP represents it.

A good counter-example would be if I need to manage a resource via an API. Maybe one request creates the resource and another tears it down. (Eg, begin a transaction and then commit it). “Using requires me to implement a new class to represent this resource - that’s OOP at its finest - whereas “defer” allows just doing the thing (and works equally well for objects)

Scoped resources are essential for avoiding global singletons. The JS/Node ecosystem acts like (and often believes) globally-scoped singletons are a recommended pattern for things like database connections.

The reality is, there’s no better option at the moment.

Virtually every other ecosystem has concluded globals are not the best practice. (At least until we return to dependency injection containers where they are suddenly cool again but I digress.)

The short answer is always, always have your own lawyer review these things.

The lawyer preparing the corporate documents is the corporation’s lawyer. That person cannot also be your lawyer. Every founder should have their own lawyer review and advise them on the paperwork.

I get the sense that ChatGPT crosses a complexity threshold where there’s no good way to describe how it works that satisfies everybody, and that’s leading to cyclical stories of the form “everyone else describes it wrong, so here’s my take.”

As a heuristic, I see descriptions falling into simple buckets:

- stories that talk about tokens

- stories that don’t talk about tokens

Anything discussing technical details such as tokens never seems to really get around to the emergent properties that are the crux of ChatGPTs importance of society. It’s like talking about humanity by describing the function of cells. Accurate, but incomplete.

On the other hand, higher-level takes happily discuss the potential implications of the emergent behaviours but err on the side of attributing magic to the process.

I haven’t read much, to be fair, but I don’t see anyone tying those topics together very well.

Where does fly.io document their per-account services limits? For example, max apps, databases, etc.

I took a quick look and couldn't find them. Do they have any documented service limits?

A google search turned up [0] which does not inspire optimism.

> ...there isn’t a limit to number of apps from a billing standpoint...

[0] https://community.fly.io/t/free-tier-limits-and-quota-needs-...

> Maybe if we trained an AI to detect genuine vs. parody?

Blockchain will solve this better