chartpath

Not to mention they will need to research how to make their models faster and cheaper to run in order to fit some margin within what people are actually willing to pay.

Are there any indications that this will be possible? Consumer hardware will continue getting better but I can't see 512GB RAM in a MacBook Pro any time soon. I'm hoping linear attention techniques plus MoE will make breakthroughs in size/compression and throughput.

Sure, and OpenAI, Anthropic, Grokkk are totally self-made and profitable.

I find Mistral Medium 3.5 with OpenCode is perfectly fine if you're willing to talk to it in a more fine-grained way about actual code. For me that's fine because even with huge frontier models I don't like trying to vibe prompt like a product manager.

Yeah same company. Clearly they haven't improved their guardrails in the last couple months.

It's never the technology that's the problem, it's the owners and operators who decide how to use it.

We should pivot the culture to one that is pro-liberal arts again. These people know how to read and write better than STEMs in general.

CS as the only path to programming was always too narrow, and often people with a broader education are better at creative solutions. With AI-assisted programming I'd argue they have an even clearer advantage now.

Some much focus on fine-tuning when it can actively make performance on reasoning and planning benchmarks worse (over a baseline of already worse-than-coin-toss).

Why not give us nice things for integrating with knowledge graphs and rules engines pretty please?

I've never had a problem working https://pgtap.org/ into CI.

I know the article title says "integration tests" but when a lot of functionality is done inside PostgreSQL then you can cover a lot of the test pyramid with unit tests directly in the DB as well.

The test database orchestration from the article pairs really well with pgTAP for isolation.

Kind of a tangent but I'm really interested in why statements like:

if you have carbon-based life forms, you will have water and CO2.

..can lead to statements like:

it is just way more likely than any other form

I totally agree on the observation, but what is fascinating to me is why a deductive statement can be considered to indicate likelihood in probability. It seems there is a bit of abductive reasoning going on behind the scenes which neither the deductive logic or inductive probability can really capture on their own.

How is wealth distribution wrong? It sounds like you are arguing in favour of inequality.

There's a third kind, which is when unscrupulous business managers or politicians use it to make decisions that they would not be capable of auditing for a rationale when otherwise required to know why such a decision was made.

It's more of an ethics and compliance issue with the cost of BS and plausible deniability going to zero. As usual, it's what humans do with technology that has good or bad consequences. The tech itself is fairly close to neutral as long as training data wasn't chosen specifically to contain illegal substance or by way of copyright infringement (which isn't even the tech, it's the product).

$10k is nothing for rich people. Just putting them in debt for it doesn't mean nothing of value was provided.

The reverse would be true where poor people could be ruined, unless the value provided is worth significantly more than the debt created, which seems doubtful.

There is a lot of synthetic content about academic subjects on YT now, and it's very low quality. I used to search for lectures to listen to while walking or driving but now need to wade through tons of enshittified spam. Even if it's reading wikipedia or other long form articles, the voices and graphics are bad.

Actually I paid for Blinkist recently and really enjoyed it at first. They have a lot of "blinks" that state at the end that the voice was synthetic and I was legitimately surprised at the quality, having not even noticed until they told me.

This seems like a good move for YT to maintain a basic level of quality (which I'm amazed can actually get worse), but I suspect it's a pretext to avoid paying out to "illegitimate creators" for commercial reasons in a way that makes them look like they care about people.

UK scored a massive own goal and are no longer part of Europe ;)

I've been wondering this myself lately.

After using RAG with pgvector for the last few months with temperature 0, it's been pretty great with very little hallucination.

The small context window is the limiting factor.

In principle, I don't see the difference between a bunch of fine-tuned prompts along the lines of "here is another context section: <~4k-n tokens of the corpus>", which is the same as what it looks like in a RAG prompt anyway.

Maybe the distinction of whether it is for "tone" or "context" is based on the role of the given prompts and not restricted by the fine-tuning process itself?

In theory, fine-tuning it on ~100k tokens like that would allow for better inference, even with the RAG prompt that includes a few sections from the same corpus. It would prevent issues where the vector search results are too thin despite their high similarity. E.g. picking out one or two sections of a book which is actually really long.

For example, I've seen some folks use arbitrary chunking of tokens in batches of 1k or so as an easy config for implementation, but that totally breaks the semantic meaning of longer paragraphs, and those paragraphs might not come back grouped together from the vector search. My approach there has been manual curation of sections allowing variations from 50 to 3k tokens to get the chunks to be more natural. It has worked well but I could still see having the whole corpus fine-tuned as extra insurance against losing context.

I thought they already didn't use input data from the API to train; that it was only the consumer-facing ChatGPT product from which they'd use the data for training. It is opt-in for contributing inputs via API.

https://help.openai.com/en/articles/5722486-how-your-data-is...

That said, for enterprises that use the consumer product internally, it would make sense to pay to opt-out from that input being used.

Just like NNs did to symbolic AI which is sorely missed to build explainable and ethical systems.

Thank you for saying it!

Search query expansion: https://en.wikipedia.org/wiki/Query_expansion

We've done this in NLP and search forever. I guess even SQL query planners and other things that automatically rewrite queries might count.

It's just that now the parameters seem squishier with a prompt interface. It's almost like we need some kind of symbolic structure again.

I can understand why that framing would be attractive, but there is no real fundamental difference when considering JSONB/HSTORE in PostgreSQL, and now we have things like pgvector https://github.com/pgvector/pgvector to store and search over embeddings (including k-nn).

Even if that were true, LLMs don't give any kind of "handles" on the semantics. You just get what you get and have to hope it is tuned for your domain. This is 100% fine for generic consumer-facing services where the training data is representative, but for specialized and jargon-filled domains where there has to be a very opinionated interpretation of words, classical NLU is really the only ethical choice IMHO.

Especially if they were hired with remote in their original agreement. Changing the location of work would necessitate some additional consideration and a new agreement.

Haha, much better response than the usual "wake me up when it's replicated in humans" any time there is a mouse study shared.

Thanks! Great points about leading with some of our own vulnerability as a way to create psychological safety. Refusing bad direction to protect the team is always hard, because it carries the burden of articulating how that protection is better for the company as a whole, which means politics and either CYA records or face saving flattery that is somewhat degrading for the manager to have to do. You sound like a good person to work with.

This resonates with me. Also in management with a similar style. However, I'd like to take a little tangent from there in case you'd want to elaborate on what it means to "get to know" people as part of the humanistic approach.

One of the ways that I've learned to practice respect is actually not really getting to know people personally too much. I've been manipulated before by "leaders" who try to use personal information like family needs or other goals as a fake carrot that is not in their power to exchange. As a result I have a pretty "no questions asked" policy on the privacy rights of my direct reports when it comes to approving any time off or facilitating whatever it is they want to do with their careers, and it doesn't require "getting to know them" beyond their own voluntary sharing in private. Sometimes it will grow into a closer relationship but it totally doesn't have to.

Maybe it's an extreme reaction to the bad taste from interacting with folks who like to pretend that work is a family. I just don't trust workplaces enough to be fully vulnerable anymore, and by extension would never demand that another employee be unconditionally trusting with their full humanity, because it is never a two-way street. When the company wants to mess with livelihoods it does not need to seek permission, so it is unfair to ask for the disclosure of personal information beyond what is necessary to carry out the work.

For just one more recent example, I fell for the advertised inclusive corporate culture by putting my pronouns on my Slack profile (having never been publicly out as a queer person at work before), I experienced weird behaviours from people that made it harder to do my job normally. So I ended up going back into the workplace closet because it's just easier to project what "regular" folks want to see in order to solve problems efficiently.

I'd love to hear your take on healthy boundaries, because sometimes I wish there could be more connection with people in the workplace besides the naturally occurring camaraderie that arises through direct collaboration and the inevitable small handful of friends with similar values. Is it possible to have a deeper connection with everyone as a general approach?

Using the same here from https://www.saaspegasus.com/

There is a tutorial up https://liveviewnative.github.io/liveview-client-swiftui/tut...

Totally agree. I once helped start a worker-owned cooperative dev agency, and discovered (naively) that getting investment was basically impossible because investors and lenders aren't interested in creative legal solutions that don't fit their mold in which they would have funded a normal company doing the exact same work. That and extensive direct democracy can lead to design by committee and factionalism, which distract from the mission or violate the bylaws and require more overt conflict (compared to the back room stuff of hierarchical companies). I also recently took a university course on corporate law and co-ops were not mentioned even once (in Canada where there is a federal Cooperatives Act that permits their creation no less), so it is very niche. See Mondragon for an example of successful worker co-op businesses.

At least unions can still exist inside the standard model of employment which everything else is built around. In a "benevolent" tech company that actually treats its people well, I have trouble seeing unions as the ideal vehicle for balancing fairness with innovation. Of course there is inherent exploitation in any employment relationship within the standard model. I would love to see an example of it working in tech though, because I can imagine a union that is equally as interested in the success of the business and prioritizes sustainable growth on top of enforcing equitable treatment as a backstop.

There has to be some combination of legal structures that can somehow represent all interests fairly without adding friction. That is unions, or unions plus other things.

For example, employee ownership trusts could in theory be very aggressive and help smooth out the continuum between periodic bargaining and striking as the main/only leverage. Similar to a share purchase plan or RSU with a pool that has a majority of voting power, or at least equal to founders and investors with all the usual tie-breakers and dilution protections.

If I remember correctly some countries (Germany?) also have legislation requiring a certain amount of employee representation on corporate boards, but I don't know if that actually makes a big difference.

At will employment is not legal in Canada.

It's not like the shares are destroyed or immediately sold. If anything, they are prevented from being sold under pain of contempt, so could increase prices by limiting the supply.