Chiawei

Engineering@Lio | Volunteer teacher & mentor @ ReDI Munich

[Chiawei @weiliddat]

the cat can stay annoying longer than you can stay patient

[Chiawei @weiliddat]

While using Claude Code in Amp I’ve accidentally used Sonnet 5.5 when I didn’t mean to (didn’t ask Puck to explicitly set model and effort). Surprise! It’s a decent model!
Just need it to be like half the price and it’ll be competitive with Sol.

[Chiawei @weiliddat]

This! Side chat and btw and forking and threads and all other tools are for me to organize my own thoughts.
The speed at which I’m forced to confront decisions is outpacing my human brain

[Mario Zechner @badlogicgames]

time and time again i fall into the trap of designing a big piece of infra with an agent, only to find out that my context window is smaller than the agent's.

i keep forgetting things that were already discussed, while clanker mcclankerface knows all the things.

[Mario Zechner @badlogicgames]

time and time again i fall into the trap of designing a big piece of infra with an agent, only to find out that my context window is smaller than the agent's.

i keep forgetting things that were already discussed, while clanker mcclankerface knows all the things.

[Chiawei @weiliddat]

@thorstenball Side chat and btw is more for me than for the model TBH. I get confused by my own meandering later on

[Chiawei @weiliddat]

Ngmi

[Joseph Cox @josephfcox]

New: Accenture has told employees it will only fund $29 per month of Claude use. Meaning that employees who often spend as much as $1,000 a month are now scrambling to find ways to keep access to the tool. A sign of the continuing crack down of AI https://t.co/Z2xswuSW5x

[Chiawei @weiliddat]

@thorstenball How do you work in a team when most things can be done solo now? What do you find still most valuable when you get together, and what form does that take (e.g. weekly formats, office hours, totally adhoc?)

[Lydia Hallie ✨ @lydiahallie]

We recently added idle compaction to Claude Code!

It now compacts long idle sessions right before the prompt cache expires (~1h). If you come back to it later, it only has to reprocess the summary instead of the entire conversation, which saves you some usage.

[Chiawei @weiliddat]

Never knew my Advent of Code puzzling skills would come in handy.

[Chiawei @weiliddat]

The difference between "Intelligent UI" in Gemini vs ChatGPT is pretty stark. Maybe also Argon vs Sol.
I had to try really hard to get ChatGPT to even create an interactive play by play to explain rolling scissors.
and I think the nose/vector for Sol is wrong

[Max Spero @max_spero_]

it should tell you something that the most bleeding edge ai-pilled technologists are not letting AI write for them

[Chiawei @weiliddat]

> Personal agents will, I think, act as a kind of structural subsidy for product quality. How would markets look different if individuals always spent at least 10 hours researching their purchases?

[Chiawei @weiliddat]

https://t.co/9UZNX4BrRC

[@levelsio @levelsio]

A very heated debate yesterday about this, which shows I hit a nerve

I asked @grok to summarize it and it said people believe I am directionally correct but a bit early

Which I agree with

It might take a few more years before everyone in the world uses AI chat apps and agents as @karpathy said

And also until most people understand what's actually possible with current and future AI models

But it's never bad to be early with these things and adapt before everyone else

One of the repeated misconceptions I see is that people think I mean people will ask AI agents to build software, but I don't, their AI agents will solve their problems by building software on the fly

With that a market springs up of software/data/etc for AI agents to pay for and connect to (like with MCP/APIs), but again the AI agents will try optimize to get the best service for the lowest price (free if possible) which makes it a cut-throat market, but possible to build a business in

Anyway, it's interesting times and I learnt a lot from the replies and quote tweets yesterday, thanks everyone!

[Quinn Slack @sqs]

You can use your Claude plan in Amp now. Free and open to everyone.

(By very very very very popular request.)

https://t.co/ndFj5NbNbI

[Chiawei @weiliddat]

@yacineMTB you use it directly?

[Chiawei @weiliddat]

@nicolaygerold orbin on a neo. nice flex. I'd like some of the label filters to be persistent. I ask Puck to label threads it "owns" as "puck", and I'd like to filter them out by default.

[Chiawei @weiliddat]

Best way I’m using GPT models is through @AmpCode Puck (Opus 5.5) to manage them.

[Nate Berkopec @nateberkopec]

I'm bearish on Dots/openai-powered consumer-targeted stuff for this reason. The basically RL'd the shit out of their models to no longer be possible to talk to. They're so good at math, computer use, but actually interfacing with them via text is nearly impossible.

[Chiawei @weiliddat]

@lydiahallie Can models like Opus just do a best guess as to what effort is needed?

[Thorsten Ball @thorstenball]

100%.

[Chiawei @weiliddat]

@liminal_warmth They might exchange a lot of information because of the context and nuance needed, but your AI PA would inform you in the best way possible if it was relevant, appropriately verbose, at the right time.

[Chiawei @weiliddat]

@liminal_warmth I think the terminology is right, just like "meat proxy", but is a passing symptom. Right now AI agent access still is very early. Imagine if everyone had an AI PA, and the AI agents just talked to each other...