Hacker Newsnew | past | comments | ask | show | jobs | submit | rufasterisco's commentslogin

https://github.com/NVIDIA-NeMo/Switchyard#routing-strategies

Looks like your great question doesn’t have an answer, but looking at the routing strategies things get even more confused, since the proposed ones tend to rely on extra llm calls to determine which model to pick.

The nice thing is that it makes sense for specific setups, less conversation oriented.

As an example, you need to classify batches of data, and have many fine tuned models. Or you need to do speed to text and need to pick which whisper to use.

You can write your own strategy, in that case an harness with subagents would be able to leverage this, picking the right model and then keeping its session sticky, but overall the lack of concern for caching points towards use cases where you do not gain much from it.


Build/vibe-code your own benchmark and find out, the work pays off and lets you try many models.

Also try fine-tuning small models and see how they perform.


Oh, I have, and even Opus falls flat. I feel like regex would be better. Oh... maybe i can have Opus create such a regex...


AI is having an impact on the legal system already.

https://archive.is/CzfHj


The article clearly states he was not on board.

There are many reasons for critizing him over things he is responsible for.

Never been a fan of mob lynching.


> The Greenland government’s statement said it “would not be proportionate” to demand the equipment be removed. An application for permission remained under processing, it added.

They could have written “ Island’s government says no denial given…” and it would have been equally true.


some small models are fast, and fine tuning can be done locally

for example in gaming context, if you need an answer below 5 seconds, they are the sweet spot


Following the same reasoning, smartwatches would have been a failure. They might be niche, but sales prove there is a market to explore/deepen.

Overall, openAI can sit on a mountain of money and bet mobile phones are the final form factor for human/AI interaction, as they have been for human/internet.

Or deploy some of that money and figure out whether that is true.

It’s also about creating an environment for those features to emerge, without being constrained by someone else platform, that could natively embed it at any time, and (probably more important) app marketplaces.

A signal for opportunity is also given by Apple’s inability to give their users Siri and Google sitting on transformers while OpenAI grabs the chatbot marketplace.

Also, anything they build does not cannibalize their other product lines.

There are lots of incentives for them to try out and figure out features on the fly, which arguably also happened with the IPhone (app store).


People were already wearing watches along smartphones. Smartwatches didn’t need a new user habit formation.

You can definitely create new habits (like people wearing ear buds for hours a day) but need a strong thesis and cost-benefit analysis that is net positive when considering alternatives.


Some people like it better when they direct the solution because they walk away with a better understanding of it.

This has emotional/psychological aspects (it feels less like LLMs are replacing you), as well as practical ones (overall complexity is bounded by what the dev brain can understand/grasp).

A dev work becomes more and more about reliability, signing off safe software with a litmus test: “I will be on to handle this code failure as if I had written it”.

All the above points towards keeping tight control over some level of abstractions and delegating others.


I find it depends at what stage I'm at with the idea - sometimes I don't want to understand it until it works, because I've wasted enough life on things that didn't do what was promised. But once I know the idea is feasible, yes I would prefer to understand the code at some level.


https://archive.is/p1ehR

The essay from Zuckerberg


Thank you! Okay yes well this manifesto is a contradiction. Yes good let's keep empowering ppl by putting AI into their hands, I love the opening. No bad we don't do that by handing you all our personal context to make these agents "work for us 24/7 to better our lives".

I'm the agent doing that in my life, that's my fucking job. I will continue to use dumb agents that I direct, because only I retain ownership and sole rights to my personal context on which my decisions are based.


I don’t understand. The model is open, you can host it and use it without giving Meta anything.

I’m not saying Meta is not eager to syphon data from users, but I don’t see how this happens via this specific product.


The intention is in black and white. The precedent of prior product design backs that up. The open models are great on their own but they are not the biggest part of the manifesto.


God, it's a bit War and Peace...


Non-archived version: https://www.meta.com/thefutureisforeveryone/ since its not paywalled.


So many em dashes...


There really wasn’t that many and this did not have the voice of written-by-an-LLM to me

And they’re not em-dashes -- they’re double hyphens.


This is unneedfully obtuse. Double hyphens are a stand-in for em dashes in most contexts.


I think the context here is that em dashes are usually generated verbatim from LLMs? I haven't seen them generate a lot of double hyphens; I suppose the post could be laundering them to double dashes, but if you're gonna try to hide LLM prose just drop them entirely? [please correct me if I'm wrong and double dashes are also an LLM tell - genuinely don't know]


My comment was trying to point on the inanity of calling out "so many em-dashes" on a post where it doesn't seem that obvious it was LLM generated. I added some of my own inanity with my remark about them being double-dashes, even though some people undoubtedly find-and-replace the em-dashes in their LLM-generated content something else to try and make it appear more authentic.

But perhaps the best approach would have been to down-vote and move on.


Isn’t that counterbalanced by the fact that the reason they exist in the first place is to be a sovereign platform?

I am assuming the reason companies switch to them is not price or tech. US cloud providers have the advantage on both.

Selling EU companies data would mean destroying trust over their main selling point, not to mention incur on EU wrath.

Feels like living one whistleblower away from doom.

I refer to fully EU clouds, parent list includes US clouds that do not need bags of money, Clouds Act in enough.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: