reedlaw

Same experience with Sonnet on low effort. It used be when I used a "table_name/id" format to reference a db record, it knew exactly how to find it using connected mcp tools. Today it failed 4/4 times (I tried the exact same prompt in 4 separate sessions and each time it replied "I don't have access to [...]"). On medium effort it got it right the first time.

Seems agents understand their own bugs now. https://github.com/openai/codex/issues/9252 has 88 thumbs up and a workaround (switch to raw mode with Alt+R) found in a comment. Codex suggested the workaround to me today when I complained about its multi-line bash command being corrupted on paste because of the two-space indent in Codex's code blocks.

> we shouldn’t anthropomorphize

Agreed, so why frame it as avoiding unpleasant people? My experience with Anthropic's latest models is that no amount of instruction can overcome their latent verbosity (nor can AGENTS.md overcome OpenAI models' reticence). I tried A/B testing several models and Opus was the clear winner, so I'm not likely to retire it just because it generates unpleasant prose. I do curate model-specific notes to mitigate some of the worst offenses.

Turns out Frigate uses visual classification only. Hard to trigger when birds are so small, unless they fly right up to the camera perhaps.

Expectations around software quality will rise in tandem with gains in productivity so that the same experience curve continues to apply.

Because agents lack human judgment. At the very least there's a need for a human-in-the-loop with agentic processes. Otherwise, it's like running a coding harness with --dangerously-skip-permissions all the time.

In my own experience as a software developer, I see no justification for the hype. Demand for software is practically infinite, and agents aren't mind readers so they'll always need workers to turn needs into prompts. Despite how much work they get done on their own, Parkinson's Law (https://en.wikipedia.org/wiki/Parkinson%27s_Law) is still in effect. You could even expand "Work expands so as to fill the time available for its completion" with "time and tokens available".

I have Bird Classification enabled in the settings, but so far haven't seen any labelled (it's only been a few days). Similarly with Face Recognition, I've only seen false positives. It does reliably detect people though (300 tracked so far). I'll have to experiment with settings such as confidence score.

Frigate, the IP camera security system, also does bird classification: https://docs.frigate.video/configuration/bird_classification...

Codex has the opposite problem. Instead of being overly proactive it's overly reticent. I have been preferring it lately, although my preference tend to switch every few months when a model or harness regresses horribly.

I'm on my third mainboard trying to combat what seems to be an AMD platform issue that causes random freeze-then-reset events. It doesn't just affect Framework. If you search AMD Data Fabric Sync Flood you'll find a lot of discussion across various forums. The open Framework issue is: https://github.com/FrameworkComputer/SoftwareFirmwareIssueTr...

I made https://constraint.fyi/ after getting frustrated on some of the harder Clues by Sam (https://cluesbysam.com/) puzzles for which the color tags aren't enough to represent everything you need to track: whether suspects have the same or opposite status, overlapping relationships, cardinality constraints (e.g. "2 of these 5 are criminal"), hypothetical branches, or where a clue came from.

I realized that all clues could be represented by Boolean formulas, allowing a solver to check whether my deductions and hypothetical scenarios were consistent without revealing the solution.

Of course the same solver can solve for individual variables, but that takes too much of the fun out of the game, so I left that feature out.

What do you mean? I'm Gen X but remember all those things. It seems like a reasonable mashup of nostalgia.

Since the Renaissance got started before the printing press, maybe you mean the press fueled it? The idea that the church found printing dangerous seems like a conflation with events that happened during the Protestant Reformation. The Catholic Church did censor works it found heretical, including unauthorized Bible translations.

One could argue the opposite conclusion, that technology helps break monopolies, but either view depends on reductionist historical readings. The truth is somewhere in between.

Good instinct. I did both. The script came out later.

I did the malware check using Claude, providing it with the same list (https://md.archlinux.org/s/SxbqukK6IA), and it did essentially the same things as this script does to verify. So either way should do the trick.

I also had an affected package installed, fortunately it was from the official repo before it was dropped and became an AUR package.

Yeah, the README looks like slop to me.

Codex is only slightly better, and that fluctuates so I switch back and forth.

I tried to create a hook that would detect when token usage was running out and write HANDOFF.md so I could switch to another agent and finish the current task. It never worked reliably. To make a hook for db queries, it would need to run before each bash call, check if it looks like a query, and then exit with a new prompt, e.g.: "Use tidewave's execute_sql_query for DB access". But then it could just ignore the prompt the same as CLAUDE.me. What if I really wanted to use bash for a specific task? The real issue is that prompts are not tightly coupled with capabilities. If we admit that, then skills are over hyped.

I'm skeptical skills will outperform training given that Opus 4.7 already ignores a 720-byte CLAUDE.md telling it to use tidewave (a Rails MCP server with 6 tools) for db queries. When I asked a new claude session about a record it called

> Bash(DATABASE_URL=$(grep -E '^DATABASE_URL=' .env 2>/dev/null | head -1) echo "ok")

even though I have in CLAUDE.md:

> For database queries, use tidewave first.

I then prompted:

> use tidewave as per CLAUDE.md. also diagnose why you failed to heed that

> ● Diagnosis first: I defaulted to shell habits (env grep → psql) instead of pausing to recall the CLAUDE.md rule that tidewave is the first-line DB tool. The trigger was "look at this record" — I should have read that as "run a SQL query" and reached for tidewave immediately.

If Opus 4.7 doesn't follow simple CLAUDE.md instructions, I'm not sure what benefits other markdown files could bring. I don't trust Opus's own explanation, but it could point to the fact that the system prompt for bash is much longer than CLAUDE.md with tidewave.

While LLM judging could be helpful, I think the tool-call assertions (https://github.com/darkrishabh/agent-skills-eval#what-you-ge...) may be the most useful thing in agent-skills-eval given that it's the only objective measure of compliance.

Why do you say HashWithIndifferentAccess shows a lack of understanding? Like many Rails features, it's a convenience that abstracts away details that some find unpleasant to work with. Rails sometimes takes "magic" to the extreme through meta-programming. However, looking at the source [1], HashWithIndifferentAccess doesn't use eval, send, method_missing, or define_method. So I'm not sure how it seems weird to someone who works more with plain Ruby.

1. https://github.com/rails/rails/blob/fa8f0812160665bff083a089...

Chinese omits articles, verbs aren't conjugated, and individual characters carry more meaning than English letters, but other than those differences I don't have the impression that Chinese communication is inherently more concise. Some forms of official speech are wordy. Writing is denser, but the amount of information conveyed through speech is about the same. There are jokes about ambiguous words or phrases in both Chinese and English. So I was surprised at your take, but no objection to your points above. Ancient Chinese, on the other hand, is extremely concise, but so are other ancient languages like Hebrew, although in a different way. So it seems that ancient languages are compressed but challenging and modern languages have unpacked the compression for ease of understanding.

Are you saying Chinese is more concise than English? Chinese poetry is concise, but that can be true in any language. For LLMs, it depends on the tokenizer. Chinese models are of course more Chinese-friendly and so would encode the same sentence with fewer tokens than Western models.

I've come to the conclusion that Hacker News is the best aggregator out there. Substack knows my interests yet gives terrible recommendations. Youtube constantly recommends the same videos or exaggerates my interest in a topic based on a few views, spamming me with related content until I watch something unrelated. The only downside of Hacker News is that its focus is narrower than other sites. But perhaps because the focus is "Anything that good hackers would find interesting" there is a bias towards things I find interesting with less noise than more commercial offerings.

Claude has trained me on the use of the word 'invariant'. I never used it before, but it makes sense as a term for a rule the system guarantees. I would have used 'validation' for application-side rules or 'constraint' for db rules, but 'invariant' is a nice generic substitute.

Then I would say this is another proof that LLMs lack intellect or ability to reason about universals. See https://michaelmangialardi.substack.com/i/186405810/test-4-p...

Why is Haskell irrelevant to the argument that LLMs can't reliably permute programming knowledge from one language to another? In fact, the purity of the language and dearth of training data seems like the perfect test case to see whether concepts found in more mainstream languages are actually understood.

This is the second endorsement I've seen today. I gave OpenSpec a shot and was dismayed by the Explore prompt. [1] Over 1,000 words with verbose, repetitive instructions which will lead to context drift. The examples refer to specific tools like SQLite and OAuth. That won't help if your project isn't related to those.

I do like the basic concept and directory structure, but those are easy enough to adopt without all the cruft.

1. https://github.com/Fission-AI/OpenSpec/blob/main/src/core/te...

Do you have examples of the task maturation cycle? I'm not sure how it would work for tasks like extracting structured data from images. It seems it could only work for tasks that can be scripted and wouldn't work well for tasks that need individual reasoning in every instance.