trollbridge

Get kids an old computer. They prefer something big they can see, with switches that make a resounding clunk, and drives that make clanking noises.

To a kid, playing games from the 1980s off a floppy disk is not a significantly difficult experience than using the very latest technology, but I suspect the former is far better for their developing brains.

Meanwhile, you can get them to learn how to type on the keyboard and learn the basics of needing to type in commands to make the computer do something, which is a lifelong skill - tapping icons on a tablet is not.

Is it? DeepSeek has already proved they’re selling inference at a profit.

A complaint from the crews was not getting the go/no-go at the same time consistently. This fixes that problem.

The purpose is to sort which customer to prioritise calling back versus putting at the bottom of the pile.

Eventually we hope to identify ones who are going to be lifetime duds.

Nobody wants to run a service like that anymore.

PJ Library has a similar program for age 0-8. (I contacted them and they’re fine with non-Jewish families participating, even though it’s aimed at Jewish families and their children.)

There is zero chance this will happen.

I use local abliterated models + paid for subs like OAI or A\ and avoid touching anything that would hit their TOS + anonymously paid for Chinese pay as you go for things in between those two.

Works very well.

Claude doesn’t push back if it thinks the thing you’re doing is old. OAI will sometimes start barking at you if it thinks you’re making a keygen even if it’s 25 years old.

Yes, RE is 100% against TOS. I’d say their TOS is poorly written

Regional African English accents, so Nigerian, Kenyan, South African?

For those customers who are still logging in with FTP…

As a precaution they reset everyone’s email passwords, presumably from an assumption people use the same ones for both.

I’m surprise they still let people use FTP.

Chat support works well. Email is nonexistent.

No, they’re still doing well.

Support is primarily via chat these days.

Pricing wise they don’t beat Herzner anymore but are somewhat competitive with Digital Ocean.

I am gradually moving everyone to my own self hosting mostly for supportability reasons. We want to react in seconds, not hours, when a customer has an issue.

Yeah so that’s why we invented tool calls, and an LLM exercises judgment when it should simply place a tool call to do simple math.

In other words… more dimensions for a bigger set of vectors?

Training classifiers is shockingly easy nowadays with LLMs. This is probably the biggest single piece of bespoke software I ship that I didn’t used to.

We have in the works a classifier for if a prospective customer will be long term profitable or not. This changes everything.

This sounds like a competency problem on the part of the GPs. They’re supposed to make these calls.

Usually you get that info at 4:30 on a Friday.

This is exactly what we’ve been doing.

For example, a client want a classifier to decide to make a go/no go decision on roofing jobs based on weather. The end result doesn’t use LLMs at all, but does use a lot of other techniques including human feedback when it makes the wrong call.

(Now the work crew gets the decision at 5 AM instead of 5:30 and the guy making the decision doesn’t have to wake up at all.)

Having lived through many corporate mergers, two small IT teams are often far more efficient than 1 big one.

More like I’m just annoyed at them not respecting robots.txt on a site that has nothing of value for them.

What’s illegal?

The site clearly says for authorised users only.

The robots file already excludes scrapers.

Yet it often is pumping out a solid Gbit of traffic.

For a while I just did markov chains. Now I use it as an experiment to use agents to generate plausible yet pointless websites.

Well, the only proper use of some of the sites is an audience of a half dozen people who know the correct login page.

It depends. Transcriptions are now confidently and plausibly wrong.

(We use this for inbound phone calls and it’s just not reliable enough yet.)

fheroes2 predates the era of vibed AI recreations which get abandoned in a week or two.

Speech to text is not exactly new technology.

They couyld have indeed presented data that only AIs would scrape or notice... which would be entirely legitimate on their part to do.

My own website spews bogus content and misinformation about almost any topic to misconfigured scrapers. I expect to eventually see it in some training sets.

Why not both?