Hacker News: wolttam

New comment by wolttam in "Show HN: Daemons – we pivoted from building agents to cleaning up after them"

wolttam — Tue, 21 Apr 2026 18:08:06 +0000

The schedule is cute.

"Complete non-determinism for everything except the schedule it runs at."

New comment by wolttam in "Meta to start capturing employee mouse movements, keystrokes for AI training"

wolttam — Tue, 21 Apr 2026 17:29:48 +0000

See: https://si.inc/posts/fdm1/

If they captured display output as well, it could be a very useful dataset for generalized computer use.

New comment by wolttam in "Bullshit About Bullshit Machines [pdf]"

wolttam — Tue, 21 Apr 2026 14:31:17 +0000

Consider cognition as simply the manipulation of information and data into something different and LLMs can do plenty of that.

Animals perform cognition.

We shouldn't confuse consciousness or agency or with cognition.

New comment by wolttam in "[dead]"

wolttam — Tue, 21 Apr 2026 14:20:24 +0000

Tells you all you need to know, doesn't it?

New comment by wolttam in "Kimi K2.6: Advancing open-source coding"

wolttam — Mon, 20 Apr 2026 18:22:56 +0000

Because sometimes (a lot of the time in my experience) third-party providers and inference engines fail to implement the model correctly in ways that are sometimes very subtle and not obvious.

Deepinfra for example is not preserving thinking correctly for GLM5.1, even though they are for GLM5. This is one of the more obvious issues that crop up.

New comment by wolttam in "Kimi K2.6: Advancing open-source coding"

wolttam — Mon, 20 Apr 2026 18:20:23 +0000

Kimi has their own subscription that works basically the same as all the others.

https://www.kimi.com/code

New comment by wolttam in "A Pascal's Wager for AI doomers"

wolttam — Mon, 20 Apr 2026 15:56:36 +0000

I also achieve better performance on cognitive tasks when I use language to first describe the problem I'm trying to solve. In fact, it usually helps quite a bit (see: rubber-duck debugging)

I'm not sure the word "intelligence" really fits what these models are doing. I do however think it's safe to say that they are performing cognition - even if it's 'simply' cognition over their provided context and even if it's entirely limited by their training set. We still have a machine that can perform automated cognition over a increasingly wide distribution of data.

New comment by wolttam in "Qwen3.6-Max-Preview: Smarter, Sharper, Still Evolving"

wolttam — Mon, 20 Apr 2026 15:13:51 +0000

Deepinfra's implementation of it is not correct. Thinking is not preserved, and they're not responding to my submitted issue about it.

I also regularly experience Deepinfra slow to an absolute crawl - I've actually gotten more consistent performance from Z.ai.

I really liked Deepinfra but something doesn't seem right over there at the moment.

New comment by wolttam in "A Pascal's Wager for AI Doomers"

wolttam — Mon, 20 Apr 2026 13:39:52 +0000

The meaning of tokens lose touch with language in the deeper layers of large language model’s neural nets.

Language is just the input/output modality.

New comment by wolttam in "Show HN: Lmcli v0.5.0 – A useable tool for LLM minimalists"

wolttam — Sun, 19 Apr 2026 22:45:09 +0000

There's a simple read-write-execute (rwx) permission model in the TUI. You can press ctrl+p to cycle through the level of tool automation. E.g. press ctrl+p once to auto-approve Read/Grep/Glob tools, press it again to auto-approve Write/Edit, and so on.

Note: the CLI flow (`lmcli prompt|new|reply`) auto-approves all tools at the moment

Streaming is of course supported!

Note: Diffs of file edits are presented for a smoother confirmation flow.

Also: reasonably sure you're a bot. Those are frowned upon here!

Show HN: Lmcli v0.5.0 – A useable tool for LLM minimalists

wolttam — Sun, 19 Apr 2026 19:58:05 +0000

I have put a lot of hours into making `lmcli` my go-to tool for all of my work with LLMs, including coding!

Written in Go for snappy performance and a lean codebase.

I wanted a tool that did not put heavy layers abstraction between me and the model, and I think `lmcli` succeeds at this. Support for more use-cases have gradually been added, but only as those use-cases have become apparent as "fundamental" to LLM usage, e.g. the agentic tool-calling loop.

`lmcli` doesn't come with a built in agent - agents are configured with a system prompt and set of available tools. The README merely provides an example of how one build a coding agent with `lmcli` (one that happens to work pretty well!)

v0.5.0 adds image support. Yes, support for attaching images to conversations in a TUI! And it doesn't feel awful?

It's not perfect, but I think it's good enough for others to find useful.

Enjoy.

Comments URL: https://news.ycombinator.com/item?id=47827162

Points: 4

# Comments: 2

New comment by wolttam in "Claude Design"

wolttam — Fri, 17 Apr 2026 15:59:29 +0000

Human inputs their idea. AI helps bring it to light faster. Human iterates.

Still human?

New comment by wolttam in "Claude Opus 4.7"

wolttam — Fri, 17 Apr 2026 01:32:25 +0000

Wow this thread has been a cacophony of differing opinions

New comment by wolttam in "The Deepfake Nudes Crisis in Schools Is Worse Than You Thought"

wolttam — Wed, 15 Apr 2026 15:49:08 +0000

I think a pretty good portion of parents would agree with a blanket "your phone shall not be seen during school hours" policy. Something that would probably need to be decided per-district

New comment by wolttam in "Gas Town: From Clown Show to v1.0"

wolttam — Wed, 15 Apr 2026 00:14:17 +0000

I think a lot of it comes down to the training objective, which is to fulfill the user’s request.

They have knowledge about how programs can be structured in ways that improve overall maintainability, but little room to exercise that knowledge over the course of fulfilling the user’s request to add X feature.

They can make changes which lead to an improvement to the code base itself (without adding features); they just need to be asked explicitly to do so.

I’d argue the training objective should be tweaked. Before implementing, stop to consider the absolutely best way to approach it - potentially making other refactors to accommodate the feature first.

New comment by wolttam in "jj – the CLI for Jujutsu"

wolttam — Tue, 14 Apr 2026 18:44:21 +0000

That makes sense, thanks for the reply. For some reason I was under the impression that there was an active drive towards a backend that was not git.

New comment by wolttam in "5NF and Database Design"

wolttam — Tue, 14 Apr 2026 18:40:53 +0000

Why shouldn’t we care about layer 2? You can do really fun and interesting things at the MAC layer.

New comment by wolttam in "jj – the CLI for Jujutsu"

wolttam — Tue, 14 Apr 2026 14:27:48 +0000

What happens when jj introduces this native repository format - break compatibility with all the popular git hosts?

If jj is so great now and works with git as a backend, it’s tough to imagine why it’s worth pursuing a native and presumably incompatible backend.

New comment by wolttam in "ChatGPT Pro now starts at $100/month"

wolttam — Fri, 10 Apr 2026 18:20:19 +0000

But Google is not a scrappy startup. They're sitting on over $100 bil in cash.

New comment by wolttam in "Anthropic's Claude Mythos isn't a sentient super-hacker, it's a sales pitch"

wolttam — Fri, 10 Apr 2026 18:14:46 +0000

And this attitude is how you get compromised in the near future.

No reason to expect capabilities of models are going to stop.