Been working on a project where I just ask all the popular AIs their opinions on stocks and compare them. Its pretty interesting, they all tend to agree with each other and have the same biases. They love NVDA and BTC for example. hallucinationyield.com/

I patched @leadtruffle to 24.22.1 before people even started tweeting about this. Running a SaaS is not a joke 😅

For this particular case, sending all your code Claude maybe compression does help at the start of a session. But I also doubt these numbers a bit because gzip just compresses your data and doesn’t by itself make anything faster unless you have big payloads relative to your bandwidth. If you think in more detail about a network request if you are sending, say 1k or 10k or 100k text chunks the network latency is the network latency. Your bandwidth is usually measured in megabits or gigabits so orders of magnitude beyond the payload size. Compression doesn’t magically speed up your latency. It does if you are sending huge payloads but normally it doesn’t for web app traffic. It actually slows down your system slightly because you have to compress and decompress. And of course it also saves you money on your cloud server bandwidth bill so that’s a more real reason to do compression for web app traffic. Super clear example: 1mb of data gzipped on a 50mbit connection only saves 32 millieconds. So it starts to matter around there. Below that it’s not important. (Ie 100kb compression only saves 3.2 ms). I guess it’s possible that Claude is making you send entire convos back and forth over the wire but it should be doing that either. It should be storing some conversations on their server and just syncing updates over back and forth. Anyway, just a nuance I think the “always compress everything” people do miss perhaps.

Wasting time on “my custom coding agent orchestrator” is just intellectual masterbation.

Another thing several people approached us about is AI agents taking payments and deposits. Are you kidding me? The average person actively hates & mistrusts AI voice customer support and you think they will give them a credit card payment over the phone. WTF.

lol we're going to get into a fun situation where AI math discoveries break all the trusted cryptography primitives but nobody trusts the next gen AI designed cryptography

What does it mean now that Luna an Haiku are cheaper than deepseek and also better AND faster? I think a lot of people saying opensource would kill closed source need to rethink their priors. Do you use Chinese datacenters instead of AWS? No? Hmm...

I wonder if physics professors will feel the same way? That's next.

Can somebody explain why this matters for a normal web app if latency between the app server + database + external APIs dominates to 99% of the request time? I use node anyway which is much faster than ruby btw.

Swapped my x86 lambdas to arm64 yesterday. So far it improved cold boot times a lot, like 30%. Otherwise it shaves off a millisecond from most requests which doesn't really matter. It is 20% cheaper too but not like lambdas are expensive.

Time to reread the dark forest and think about it in terms of network security and ai tools.

This is very interesting bc if you are home service company starting to get bombarded with email from people like this @leadtruffle is good at doing a bunch of ai mediated automations from email.

I've been using GPT 6.1 all day today on 3 different long parallel tasks and my usage went down only 2%. I already like this model much more than Astra btw, way more reliable and trustworthy.

Some math on bandwidth capacity of new v3 starlink satellites on starship cost curves vs the old v2 satelites + falcon 9.

The fact that Astra can now super reliably do all my annoying bookkeeping tasks like download receipts and uploading them to Mercury correctly is huge in itself.

We have a customer that has now cancelled twice tried out competitors and come back 🤣. I'm not sure if that means our product is good or just sucks less and everyone has a convoluted product in this space.

I've seen things you people wouldn't believe. Attack ships on fire off the shoulder of Orion. I watched C-beams glitter in the dark near the Tannhäuser Gate. All those moments will be lost in time, like tears in rain

Once a month I go check on instagram and I notice the slop reels on there are better and more addictive. This time, I gotta say, every AI generated reel was a straight up gem.

uuuuhhh long softbank?