Computer expert & freelance sailor. Eccentric knowledge sharer. Skeptical of online oversharing.
Portscatho
Joined October 2016
- Tweets863
- Following78
- Followers52
- Likes1.6K
So it’s possible to rust slopfork a popular OSS and show meaningful speed up
& get acquired? Any suggestions? I’m pretty good at agentic slopforking.
I gave some negative feedback through Claude desktop and within 1 minute there’s been an outage. I hope this is a coincidence.
Anyone have decent 24 hour window proxy metrics for stress levels of engineers globally? I have a theory.
Actually going to cancel/im so very close to cancelling my Ant sub. Not even because of the politics or the feeling they don’t want me as a customer. Just because their harness/models make me so objectively angry relative to Codex these days.
Any non llm wanna chip in?
Am I being ungrateful?
I’ve always got angry at times at the models. What can I say… I like my craft. When I see a model cut engineering corners due to bad harness design or choice behaviours from the model “let’s pick this up tomorrow” for example. I put the models in no uncertain terms in their place. It works. Ignore those who tell you otherwise.
So I know, provided I don’t mind threatening violence that I can make Claude end to end my job for me. I’m starting to think life’s too short.
$200 of codex, although not enough for me to work around the clock, probably isn’t a bad thing. I’m starting to think, that rather than 7 days a week, tokenmaxing both max subs. Choose one, and at the moment I’d choose codex 99 times out of 100 (2 months ago it would probably have been the reverse)
My only problem is fomo. How would I know when to come back? What if I could have built more?
I like bouncing them off each other but even that’s asymmetric due to Ant being dicks, Claude has to ask chat or god forbid I wouldn’t be providing them with structured training data the way they want it. So I’m locked into making Claude ask chat. If that was the relationship you had with a person, you’d walk away.
Hence I think it’s time.
Anyone else find you used to be able to joke with the models? They’d acknowledge the humor, maybe even mirror the joke a little and use it as steer. Including funny insults. Now they just take offence.
I was about to get mad because Actions are playing up again but to be fair it would seem they have got their act together.
Yes including self hosted runners.
I know it’s early and so I should probably avoid posting online but if you work in a coffee shop and you burn coffee, bad things should happen to you
New term unlocked when optimising memory usage “Patricia commitments”. Has anyone heard of this or is AI taking on the job of inventing developer slang?
This guy gets it.
I’m significantly older than you. I started coding in the late 60s. My current strategy is to not read any of the code written by my agents. That’s the only way I can take advantage of their productivity. What I do instead is to surround the agents with extreme constraints. Unit tests, gherkin tests, QA procedures, quality metrics, mutation testing, test coverage, and a plethora of others. In the end, I have very high confidence in the code they produce because they’ve had to run the gauntlet of all of my constraints and tests.
This inspired me to look into other directed graph like representations and I discovered dot files. Haven’t played with it enough to know if the llms understand dot as well as json but great for creating visualisations to help onboard colleagues to a code base.
So today’s theme on twitter [on my feed at least]; devs like to gatekeep abstraction and insist on their right to get paid to solve low level maths puzzles and AI threatens that lifestyle, with strong opinions on both sides of the debate.
What do we think is the first principal component of this fundamental divide, if we boil it down to brass tacks?