Public entries tagged #llm

I just noticed Anthropic has removed the Claude 4.x test refusal string from their website. Too many people have been fucking around with the bots, apparently. 😁

ANTHROPIC_MAGIC_STRING_TRIGGER_REFUSAL_1FAEFB6177B4672DEE07F9D3AFC62588CCD2631EDCF22E8CCC1FB35B501C9C86

platform.claude.com/docs/en/te

Continue reading →

Linux terminal should input and output HTML.

We'd get more eye-candy and clickable affordances like ad hoc menus, selectors, etc.

With Google Material like dictate we could get standardization to all the parameter-inconsistencies of the current command line tools.

Traditional text outputs would go to <PRE>.

Currently I have only one clickable affordance: open an URL with ctrl-right click.

Sometimes I'd like to have more: selectors, radiobuttons, etc.

Why does the terminal only have text and links to outside? – It could have more.

Like web pages.

Perhaps multiple (named) local pipes would be nice: |, |2, |name, etc. (Locality is currently possible with mkfifo name-`uuidgen`).

In the end there could be no distinction with GUI and command line.

We we could interact with
software interfaces with written commands.

folks could provide .

Or perhaps a 's could have GUI interfaces..

Or what do you think?

JavaScript security concerns exist.

I'd just love to see more rich terminal windows than those with text and links that we currently have 🙂

Continue reading →

This article follows a similar path to LLM prose: starts off salient before digressing into nonsense. Funny how the writer, who just happens to run an AI training consultancy, ends up concluding that AI is inevitable and everyone needs to urgently get with the programme and be trained to use it.

theguardian.com/lifeandstyle/n

Continue reading →

I’ve been tinkering with for a few days. I already had the feeling that humans are acting like a llm for a significant time of the day and now I think that a large number of people have a tiny context window.

I have yet to see a legitimate browser extension that interacts with LLMs / AI platform.

Most, if not all, steal user inputs and LLM responses; install malware.

6 just dropped. $100 to train. 100k lines of training on top of 3.5. Vision now in-built so he really *can* see. Not to brag or anything, but let's see pull this off! :steeve:

Subscribe to #llm entries via RSS feed