legendofbrando

I think this is precisely right. Like 100% correct. No notes. Exactly nails it.

In similar news: my left hand acquired my right hand today in an all stock deal valuing the combined hands at $1T. Praising the announcement my arms noted on the deal: “With these two hands now together, there’s nothing our combined fist of might can’t do.” Competitors, my left and right feet, declined to comment on the merger but are said to be in their own separate talks about a deal.

Huge plus one to this. We undervalue when to bet on data and when to be comfortable with gut.

@simonw you’ve been awesome all year; loved this recap and look forward to more next year

This is so cool!

You because you didn’t have to come out of pocket the high cost of a down payment, handle maintenance costs, deal with the risk of mortgage default if you lost your income, deal with the possibility (and thus hold insurance for) of liability if someone is injured and sues, or lose out on the opportunity cost of money spent on the house versus what else that money could have purchased. To say nothing of you don’t know what their interest rates are, how property taxes or cost of home owners insurance may have changed (in the U.S. your escrow payments change often if taxes and insurance change) or how liquid the house is (ie could they sell if they needed to and at what discount to the market value).

Oh, and don’t forget that they’re paying interest on the mortgage and that - when adjusted for inflation - the actual increase in value versus what they’ve paid in in mortgage interest over the years is probably far less than the non adjusted gains it looks like.

It’s pretty easy to demonize owner landlords when you’ve always been a renter because you think only about a monthly payment. I’m not going to tell you that’s it’s a relative luxury to be fixated on one simple payment each month, but it’s also not the case that owning a home as an individual is some kind of pot of gold.

An owner that values their tenant and keeps their rent flat isn’t a saint. But they’re also doing a good thing in a time when they could be - by account of this thread - exploiting people for as much as possible. We don’t need to order them a parade, but it might be worth broadening your understanding of what the cost of a home is before you blanket assume they’re worthy of scorn.

The finding on simpler prompts, especially with GPT4 tracks (3.5 requires the opposite).

The take on RAG feels application specific. For our use-case where having details of the past rendered up the ability to generate loose connections is actually a feature. Things like this are what I find excites me most about LLMs, having a way to proxy subjective similarities the way we do when we remember things is one of the benefits of the technology that didn’t really exist before that opens up a new kind of product opportunity.

The point of the punishment is to tell you how much we don’t want you to do it. You get life for murder because we really don’t want people killing people. Fleecing them out of money is bad, but we can (literally) print more of it. Once someone is dead we can’t bring them back.

Your analysis is flawed because you’re looking at the problem from the vantage of collective outcomes. A murderer is one person who is making a decision to take an action that can end one or many lives. You want them to weigh that heavily as an absolute cost (I will go to prison forever if caught) not some relative cost (well if I just murder a little then I’ll only get a little punishment).

Don’t do the cleaning and leave a medium review that politely says you were not informed of cleaning policy beforehand. Host has no way to penalize you (outside of a bad review) and you can respond to their review (if they criticize you) saying you were not informed at booking. Keep your review and response cordial and any future host evaluating you will assume that the past host was weird. There’s no way anyone can verify the claim by the other host that they articulated the cleaning policies.

I’m not equivocating AI to these inventions, but imagine if this dour, pessimistic outlook greeted electricity, the combustion engine, the car? New things aren’t immediately valuable and come with all kinds of tradeoffs. We would hardly recognize our lives without these things now and - even though they come at great cost - no serious person would have given that progress up.

It’s fair that “we” (the collective human race) have been too optimistic about new things and too sanguine about the cost. That’s why climate change is as bad as it is. That’s why having everyone plugged into social networks makes us sad.

This constant stream of pessimism that started in the 10’s for everything (tech, any institutions, the world) etc is just so..small seeming. Complaining like this makes none of these things better and the ideas expressed in this piece are far from unique. So as criticism goes it fails to even meet that definition.

So why even do it? To feel cool I guess to other people who think being pessimistic about everything is cool?

We’re coming up on a decade plus of this prevailing dour attitude and not a single bit of it has helped forestall anything. Maybe we try a new approach where we treat new things as things to be improved when they’re at the beginning (and therefore most able to be shaped). Maybe we try being cautiously optimistic?

I do hope you’re recognized.

One piece of advice, I found myself this person once - it’s the road to burn out city to be the critical path answer to everyone on the team’s challenges. You sound like someone who is very generous with your time and support. In my experience once people have found a critical path there is no point at which they “stop.” This isn’t because they’re trying to hurt you, they’ve just found the answer so to them it doesn’t seem wrong.

It will take even more of your time, but you ought to consider practicing giving less answers and asking more questions of those who seek your help to try to help them unpack the issues themselves. It will feel more tiring at first, but you’ll gradually help the others learn and also create a small bit of friction that will encourage them to try their own solution or two before seeking you.

Management asks are separate/ they can actually reward you with compensation and promotions for this extra work. But the team asking for help won’t stop when you get more comp unless you start teaching that you’re not the answer.

Still there for me. Kept meaning to spend the boot up time to migrate to Swift but have always been more keen to just get going. Now finally feeling a bit “forced” as I want to tap into Swift only things like widgets and live activities.

Having done some of this myself, I’m curious your results on fine tuning vs embeddings. I’ve found the latter much more performant, but perhaps I’m thinking about fine tuning wrong.

This is one of the best pieces I’ve read that articulates what it’s like to work closely with LLMs as creative partners.

Surely one answer is to train (or aggressively fine-tune) a new model that doesn’t (or refuses) to produce these outputs and then - as exists already, augment that model’s understanding of copyrighted material by having it Bing/Google search as a RAG process that requires the end user to log into accounts at the New York Times (and other accounts) with their paid sub. This broadly replicates the process a person could do today when they read the internet and summarize it while paying rights holders.

Expensive to do but hardly the end of Generative AI or OpenAI should that be the difference between having a business or being sued out of existence. Never underestimate people who have a clear economic interest especially when their own existence is at stake.

Agree. My view is we’re in the Napster moment and someone is going to invent the iTunes Music Store. Language models are a distribution mechanism for knowledge content—- in many cases more efficient and useful than the originally packaged materials (akin to how downloading a single pop song is greater than buying the album). It feels clear this is where we’re headed (verified, compensated content delivered through a new mechanism); this lawsuit is like the RIAA v. music sharing and the question is just if the current players in AI make it through or if someone else will come in and do iTunes.

What’s the benefit of llamaindex over just storing documents in chroma and using chroma to query? I’ve done the latter and trying to understand if there’s a performance gain to the former?

Most people who dislike political people are failing to place the blame on the right actor. Sounds like this engineer has recognized how to get rewarded at their job and has adjusted their behavior to fit. The actual root problem is that this reward system fails to recognize the quiet everyday progress required to make something great.

Yes to this. Too many people complain about politics as if it is some “outside” force and not simply implicit in the way that human beings organize diversified interests. Since each of us can only embody our own world view, politics is inevitable.

I understand that what people typically mean is people who are “too political” where the pushing for interests stops being in service of some goal and starts to become its own point. But in trying to avoid the appearance of that, too many people fail to engage in any politics and in doing so actually cause more harm than good to what is important to them by missing the important point that it’s incumbent to speak up for things you need and believe in because people who allocate resources may have insufficient knowledge or insufficient time to dedicate to knowing your work that intimately.

When I try to create an API key it says that "We are sorry, but you do not have access to Early Access Apps" yet my domain does allow access to early access apps....

Crypto thinking is always basically some version of how modern finance is broken, yet the solution is basically “hey consumers, whenever you buy something you’re at risk that you could have no recourse.”

Of course individual merchants dislike cards, they create a wedge between consumer and merchant -/ but that wedge also enables people to freely spend more than they’ll ever extract from falsifying things, thereby benefiting all merchants.

Crypto logic ignores that the modern banking system is actually far more logical than it sounds at first blush when you just take the perspective that not everyone should be trusted all the time.

You me and everyone else on this thread are here getting value out of it. That’s more than the number of people who get value out of crypto.

Deeply appreciate the outreach- just sent a note and mentioned your name. I’d gotten a note that you all would have update on my api access within a few weeks so sent that along so the support team has the context

Same experience on my side; they're uninterested in actually running the business it seems.

I mean, that would be predicated on it actually being possible to get access to and use their models...which in my experience is basically a limitless void. Meanwhile I spend hundreds of dollars a month with msft/oai.

I would love to use their API but I can never get anyone to respond to me. It's like they have no real interest in being a developer platform. Has anyone gotten their vague application approved?

In my experience if you give a time estimate, people make that the new metric to optimize to. So they spend 4 hours creating something that looks like 20 mins.

Most articles like this focus on what is lost for employees at the whim of a careless manager, but I think they often miss a point: the team researching X random thing means they may learn other knowledge and their overall understanding of things goes up which - though inefficient - may have value later.

Moreover, work is broadly inefficient because human coordination is inefficient. People bring their own stories and narratives and goals. Hence the reason the 20min ask is now the 4 hours to make it look like 20 mins. Treating it like something that can be optimized misses the point of it.

Building specialized agents is a great way to achieve what humans do through teams and organizations. For individual developers the benefits are dubious, but for ChatGPT, it’s obvious that the end game is letting developers/people build and humans curate (through downloads, ratings, usage) the most interesting agents and then setting ChatGPT up to be grand central to use that knowledge to hire the best agent/teams for a job (or putting them against each other and presenting pros/cons).

To that end, the ease of building is key: lots of folks would be happy serving their little niche with no monetary gain (see open source software) which, for non-developers has never really been possible outside of social media posts/influencers.

Maybe it pays off, maybe it doesn’t but generally the strategy feels sound.

This is not hate. Langchain was a delightful way to get used to concepts and then it was easier, faster and better to reconstruct the elements I needed myself without having to dig into LC docs to try to figure out why X was acting weird sometimes. I wouldn’t recommend for anything more than that.

Would have been fascinating to see what names like Apple would have looked like on this list. Macro the trend is correct but I’d be curious to see how different it would look for what is commonly regarded as a more trusted brand. FB, Google and Amazon are sort of the “worst of” players in the space reputation wise.