How much effort did you spend to make the local setup work well?

Local models are not plug and play, you need to adjust settings to make them behave intelligently.

On my side one model looking super dumb right after install ended up being 'agentic' after using the right chat template, fixing how tool calls are setup, fine-tuning plenty of params values, and adjusting the whole setup to my own hardware. Claude.ai (free) was great at helping me sort that all out.

The AGENTS.md file can be found in the official .tar.gz source code distribution.

Comments URL: https://news.ycombinator.com/item?id=50165158

Points: 4

# Comments: 2

I used the settings recommended on huggingface/unsloth's model page + whatever suggestion from various LLMs - didn't research too much myself which setting did what.

Did I miss something, is it possible to have that model be reliable without thinking?

Comments URL: https://news.ycombinator.com/item?id=48995806

Points: 2

# Comments: 0

Comments URL: https://news.ycombinator.com/item?id=48983601

Points: 3

# Comments: 0

> Quantization-Aware Training (QAT) [...] allows preserving similar quality to bfloat16 while dramatically reducing the memory requirements to load the model

What failed with Mistral?

Which anti-AI regulations are we talking about, and don't these apply to any solution distributed in the European Union, hence also to American ones?

Yesterday, gemini told me to run this:

    echo 'export ANDROID_HOME=/opt/my-user/android-sdk' > ~/.bashrc

A few minutes later, asking it to create a .gitignore file for the current project - right after generating a private key, it failed to include the private key file to the .gitignore.

I don't see yet how these tools can be labeled as 'major productivity boosters' if you loose basic security and privacy with them...

Comments URL: https://news.ycombinator.com/item?id=46486269

Points: 10

# Comments: 0

I'm not sure of what "production ready" is supposed to mean here, but the demo image is not optimized, `optipng` command decreases its size by 53.21%.

It might be worth considering a feature to time/schedule each flow's animation, rather than having them run in an infinite loop, all at the same time.

UX feedback:

* The animation and the whole interface are sluggish on firefox/linux. There's about 1 sec delay after each action (like clicking on an option). * The site's CSS does not load on an old version of Chrome - v90 - (and the chart and animation don't either).

Comments URL: https://news.ycombinator.com/item?id=46013097

Points: 3

# Comments: 0

Bug: I tried in my area in the Canary Islands and all the places were off, sometimes in the middle of nowhere or even in the sea.

Also, in small villages, we don't necessarily have a town hall, a library, etc (within selected radius), but the game asked to pin these.

Well thought, sophisticated ways of modeling data for analytics purposes -using established approaches - are being replaced by just pulling data from the data sources - with barely any change in the source structure - into cloud data platforms.

In the past we used to model layers in a data-warehousing infrastructure each with a purpose and a data modelling methodology. For instance, an operational data store (ODS) layer, integrating data from all the sources, with a normalized data structure. Then a set of datamarts, each of them containing a subset of the ODS content, in a denormalized format, focused each on a specific functional domain.

We had rules, methods to structure data in order to get performant reporting, and a customer orientation.

Coming from this world, it seems like data governance principles are gone, and it feels like some organisations use the modern data stack same way as each analyst would be doing their own Excel files in their own corner, without any safeguards.

Some alternatives:

  * don't have web analytics at all
  * self-host a Plausible Analytics or other open source analytics solution
  * use the data from server-side access logs (for those using nginx, apache or other similar solutions)
  * use Vercel web analytics' free tier (relevant for kanadojo which appears to be hosted there) - more privacy friendly than Google Analytics.