solumos

I’m in the tiny minority.

devs just writing their own random queries.

I've spent a lot of time writing my own random queries. I don't know that I've ever written a stored function.

Small-language models also use GPT. And I agree it's not "LLM vs RL" and that they're orthogonal, however, RL has features that more clearly resemble intelligence.

Google's definition is sufficient[0]. I'd also go so far as to say that LLMs aren't even AI, but that can be a bit pedantic. I align with Richard Sutton's reasoning, in that LLMs (and GPT models more generally) aren't "intelligent" in the same way that RL models are[1].

[0] https://cloud.google.com/discover/what-is-artificial-general...

[1] https://www.youtube.com/watch?v=21EYKqUsPfg

I think it's a bit nuanced, and maybe poorly explained by the original author, but to me, "left-pathers" are always "move fast and break things" to the point that whatever they build really only works as a throw-away prototype, and the effort to architect sensibly is minimal.

"We don't really need to use REST, we can just create some endpoints that have undocumented side-effects. We don't need to abstract vendor calls into a separate class, we can just implement that functionality directly in our endpoint code."

These sorts of decisions aren't actually materially faster, they're just lazier. And maybe that's "a sprinkle of QC"? But it's a lot of unforced errors that don't really save time to implement, and also create a lot of problems later on.

On the other end, with the "right-pathers", you can have people that really try to over-engineer at any opportunity. This is sort of typical of people who have worked in much larger teams. This can mean building out a k8s cluster when you're still a team of 2-3 people, splitting into 10+ microservices, deciding to use Kafka when a simple queue system would work, building out in-house load balancing for dubious reasons, etc.

The middle path is really something that resembles the "Best Simple System for Now" — when I've done this, I think about how I can solve a problem and not have to rebuild it entirely within 12-18 months.

I’ll go on record and say that AGI will never happen (in the next 50 years). I think that’s also the timeline for white-collar job automation that requires critical thinking.

If this is true, who is the EU-equivalent of Palantir?

With integers/floats, he's saying it's not opinionated enough. Anything other than integers with minor-unit precision, unless you have a very good reason, is a bad idea. So "floating point is almost a bad idea" doesn't go far enough, and the other alternatives are presented somewhat equally.

The FX critique is saying that it's glossing over a lot of the complexity. I'd say the same is true for the treatment of DE ledgers, and it borders on bad advice (e.g. "Balance is never stored. It’s derived from the movements of money.")

Yeah, if blockers are coming out in 1-on-1 meetings, that’s a really bad sign

Without regulation, yes. Brokers (i.e. scalpers) will buy up tickets to events and take all of the risk off of TM’s plate, and reprice however they’d like. ~80% of tickets in the U.S. are sold this way. Stubhub has done a great job of lobbying for this since their existence depends on ticket brokers.

If you want to see what that looks like, I one-shot a browser with Claude that does it[0]. Docs pages are early adopters to this[1][2], so that AI agents can better handle tasks.

[0] https://github.com/solumos/md-browse

[1] https://docs.stripe.com

[2] https://vercel.com/docs

Something different that other folks might not have thought of: Robust multi-environment infra deploy scripts that leverage terraform + AWS SSO

I've found that converting stuff that's previously been very ops-cli heavy into very detailed skills has worked really really well.

I use Claude Opus 4.8 + Conductor as my daily driver

No, requesting "Accept: text/markdown" in the headers and returning markdown is the more agreed upon standard at this point.[0]

[0] https://acceptmarkdown.com/

Adults should aim to do between 560 and 610 minutes a week of moderate to vigorous physical activity

Observational study

Can we stop doing this please?

I see nothing advocating against requiring KYC for user wallets and continuing to do wallet screening (which is the norm)

I agree — I don’t see the connection between the items mentioned on the page and flat out AML/CTF evasion as the title suggests. If I’m wrong, someone please connect the dots.

This is such an important step forward as we start to understand the 2nd order implications of AI and how it will change UIs in the future.

We used to have to allow + review 3rd party plugins for software so that people could customize it, but when the cost of development is near-0, we can simply hand over the development reigns to customers.

No meaningful benefit? Tell that to the people in emerging economies who now have access to digital US dollars.

Bitso processes 10% of US-MX remittances. 12% of people in Argentina use stablecoins. The shift is happening in real time.

“Horrible” and “low cost” are contradictory for many consumers

Polymarket is simply an exchange for these sorts of “contracts” and the results are verified by a separate entity (it’s a DAO, which of course can be manipulated, and was the subject of controversy due to some Venezuela invasion-related “market” resolutions)

There are three-sided marketplaces, like DoorDash, etc:

Restaurant, courier, customer

"Focus on one side first" means get in your car and start fulfilling orders.

Plenty of people, but how many companies?

Plenty of people bake bread from scratch without a mixer, but few (if any) bakeries do.

No no, MemPal is a memory system, not an LLM

I think it's just explaining the difference between comms/collab being a supporting thing vs the only thing? It doesn't seem intended to be deep to me, but it's a little verbose.

My ISP (Spectrum) is also blocking it as suspicious...it's kind of funny that they haven't migrated the blog to SSL

It's definitely not 140ms round-trip. Issuer processors typically have a ~2.5s window for approval.

In practice, the POS sends a message to the acquirer processor -> hits the network -> is sent to the issuer processor, and back again.

https://medium.com/wharton-fintech/the-anatomy-of-the-swipe-...

Yes, just like Jony Ive has been doing

It can make/take phone calls[0], but they need to be prompted on the nature of the call, the data they need, and how to collect it. They can also output the results of the call via API. An AI agent from Masterworks recently called me using this technology.

[0] https://vapi.ai/

You generally want to IPO (and fundraise in general) at the most favorable terms possible — not when there are market headwinds.