bnfcl

6 giorni fadiscuss

How strange, works fine here. What OS/browser do you use?

6 giorni fadiscuss

In an attempt to uncover biases and default choices AI makes, I asked 100 AI models, 100 simple questions, 3 times each.

I am not sure what this tells, but I found it interesting that most of the models were very much in agreement. With some outliers like Mistral and Llama on some questions.

I even made a benchmark, ConsensusBench, to measure how aligned they were: https://www.modelbias.ai/consensus-bench

bnfcl

6 giorni fadiscuss

Agreed!

bnfcl

7 giorni fadiscuss

Thanks! Because I think that models are becoming better at creating SVGs in general. If you look at Claude Fable 5 and Kimi K3 for example.

In my tests it did create bicycles the most, but this is just a general bias I believe, as tested here: https://www.modelbias.ai/prompt/transport

bnfcl

7 giorni fadiscuss

This is funny, I actually did a similar experiment just yesterday.

Looking for evidence of the same, but with another twist: checking if the models would choose to create a pelican on a bicycle, if no specific bird or method of transportation was specified.

My version of it: https://www.modelbias.ai/pelican-on-a-bicycle-test

bnfcl

7 giorni fadiscuss

Circumstantial tacit knowledge can probably be approximated through examples and feedback. But that is not quite the same as truly acquiring it. Bridging that gap would likely require a fundamental breakthrough. And sure, there is no objective right judgement, but that is the same for a lot of things the models already do.

bnfcl

7 giorni fadiscuss

The bottom line of the article:

    Intelligence is being automated. Judgment is not.

This is true. Like with a lot of similar takes on AI, tacit knowledge is not easy to add to the AI’s training set.

bnfcl

8 giorni fadiscuss

Must be a lot of prompt injection potensial in AI evaluating user generated content like this.

bnfcl

8 giorni fadiscuss

Exactly, tacit knowledge is not easily added as AI training data.

bnfcl

8 giorni fadiscuss

Understanding the fundamentals of anything will always be a valuable skill, and that only comes from hard work and experience. And much of the experience lies in tacit knowledge, not easily added as AI training data.

That, combined with knowing what to build, and more importantly what NOT to build, I don’t think AI ever will take away from us.

bnfcl

8 giorni fadiscuss

Quote of the main point in the article:

    In other words, the hard part moves from recall (“How do I write this?”) to judgment (“Does this actually make sense?”)

This is very true. But to evaluate if it makes sense, you first need experience writing the code. I am glad I learned software development over 15 years ago, and not today. AI is a super power, but without the experience to guide it, it can go horribly wrong really quickly.

bnfcl

9 giorni fadiscuss

Had the same though. The gap between open-source and the frontier is closing in, especially with Kimi K3, but that is like >2T parameters. The Gemma 4 and other models you can actually run on an average Mac, is not in the same league.

bnfcl

9 giorni fadiscuss

Pretty shameless copy, but also shows how easy it is to do so. OpenAI and Anthropic had first-mover advantage, but easy come, easy go.

bnfcl

11 giorni fadiscuss

The enshittification just keeps getting worse, where will it end?

bnfcl

11 giorni fadiscuss

Humans, huh

Submissions

[untitled]

Show HN: What's AI's go-to, public or private healthcare?

[untitled]

comments