rectang

6 ms·

121.
by rectang 5mo ago
I feel like I'm using Claude Opus pretty effectively and I'm honestly not running up against limits in my mid-tier subscriptions. My workflow is more "copilot" than "autopilot", in that I craft prompts for con
122.
by rectang 5mo ago
The broader theme of antagonism to Black success motivating the thoroughness of the destruction is a common observation about Tulsa.
123.
by rectang 6mo ago
Seeing another one of these breaches had me returning to look at local-first software. https://lofi.so I feel like if we're going to make progress in preventing wholesale data breaches it will be through architectural inno
124.
by rectang 6mo ago
I see lots of speculation that Anthropic needs to cut usage because they are compute constrained. If that's the case, will they be focusing on reducing compute costs for their models? From what I can tell Opus 4.7 is more resource-int
125.
by rectang 6mo ago
Who says I believe it? ;) I'm just choosing to focus on the substance of the argument itself, which I think is risible regardless of who makes it and why.
126.
by rectang 6mo ago
I'm unpersuaded by the assertion that closing the source is an effective security bulwark. From that page: > Today, AI can be pointed at an open source codebase and systematically scan it for vulnerabilities. Yeah, and AI can also b
127.
by rectang 6mo ago
I've started messing with this and the experience seems pretty similar.
128.
by rectang 6mo ago
My experience is similar. Opus, especially Opus 4.5, understands my intentions better even when poorly phrased, and more consistently follows my instructions to do only what's necessary and no more. As far as I can tell, the distincti
129.
by rectang 6mo ago
Welp. I already added a $20 Claude Pro subscription to complement my $10 Github Copilot Pro subscription and $10 DuckDuckGo Plus. That was partly to show support for Anthropic after the OpenAI/DOD episode, but also because I've
130.
by rectang 6mo ago
I suppose that my generalization was too broad and that LLMs can be either good or bad at writing tests depending on your workflow and expectations. I'm closely supervising the LLM, giving it fine-grained instructions — I generally und
131.
by rectang 6mo ago
44% of uploads are probably not created by 44% of "artists". The core of people who are looking to exploit the system are going to be good at gaming the recommendation algorithm — they're specialists in it solely for the mon
132.
by rectang 6mo ago
> I recently deleted a whole bunch of automated tests because if the AI is going to write most of the code then I should test it to make sure it's good! ?? You say you deleted the tests, because you "should test it"? Th
133.
by rectang 6mo ago
Are the benchmarks being used to measure these models biased towards completing huge and highly complex tasks, rather than ensuring correctness for less complex tasks? It seems like they're working hard to prioritize wrapping their arm
134.
by rectang 6mo ago
I've used Sonnet a lot. It is not as good as Opus at understanding what I'm asking for. I have to coach Sonnet more closely, taking more care to be precise in my prompts, and often building up Plan steps when I could just YOLO a
135.
by rectang 6mo ago
https://github.blog/changelog/2026-04-16-claude-opus-4-7-is-... > Over the coming weeks, Opus 4.7 will replace Opus 4.5 and Opus 4.6 in the model picker for Copilot Pro+[...] > This model is launching with a 7.5×
136.
by rectang 6mo ago
For my working style (fine-grained instructions to the agent), Opus 4.5 is basically ideal. Opus 4.6 and 4.7 seem optimized for more long-running tasks with less back and forth between human and agent; but for me Opus 4.6 was a regression,
137.
by rectang 6mo ago
For now, I'm planning to stick with Opus 4.5 as a driver in VSCode Copilot. My workflow is to give the agent pretty fine-grained instructions, and I'm always fighting agents that insist on doing too much. Opus 4.5 is the best out
138.
by rectang 6mo ago
Sounds like promises are worthless and only capabilities matter.
139.
by rectang 6mo ago
Cassettes are a pain. Head alignment is extremely important for analog tape fidelity, and it's always off for home recordings. With pro analog tape recordings (e.g. 2-inch 24 track, half-inch 2-track), you record alignment tones onto
140.
by rectang 6mo ago
To be clear, I'm not sitting in judgment of you or any of the other spoiler trolls, not back then and certainly not now. This is an instance where the Potter-philes couldn't fight back, and to my mind that's inevitably going
141.
by rectang 6mo ago
There's never going to be philosophical consensus on the "good/evil/social/antisocial" debate because the human impulse to self-justify and believe that you're the "good guy" is extremely power
142.
by rectang 6mo ago
A mild reminder that humans tend towards inflicting pain on their fellow humans when there are no consequences.
143.
by rectang 6mo ago
Selective incompetence for fun and profit: Pricing mistakes which make the supermarket money are unfortunate but low priority. Pricing mistakes which cost the supermarket money must be fixed immediately.
144.
by rectang 6mo ago
If UTF-8 represents the triumph of a design prioritizing backwards compatibility with an existing standard (ASCII) to facilitate a transition, then IPv6 is the cautionary tale of a design which could have made the transition simpler but d
145.
by rectang 6mo ago
The Coltrane changes are great, but on the scale of other harmonic innovations over the years inside Jazz and other traditions. They aren't analogous to Einstein. What makes Giant Steps so amazing is the sheer speed at which those cha
146.
by rectang 6mo ago
> one or more teams of researchers in the field going dark If the intelligence community is going to nab the first team that has a quantum computing breakthrough, does it actually help the public to speed up research? It seems like an
147.
by rectang 6mo ago
> https://visidata.org/ai Thanks for that list of levels, it's helpful to understand how these things are playing out and where I'm at in relation to other engineers utilizing LLM agents. I can say that I fee
148.
by rectang 6mo ago
Oh, IOT shock collars. Hope this company takes security more seriously than most IOT players, or some sick people are going to take hacking into these devices as an opportunity for animal cruelty. :(
149.
by rectang 6mo ago
> never underestimate people's power to ignore. One of the hardest things to do is to put yourself in the place of those you see as villains and recognize that they generally see themselves as heroes. The human capacity for self-
150.
by rectang 6mo ago
I've just bought my copy. Best-sellers don't sell that many copies in the absolute sense. From what I can tell Careless People has sold around 200,000. Moving the needle just a bit is worth it.

More ›