omeze

Isnt Groq pivoting to the IaaS/SaaS model because hardware channel sales is hard and its easier for everyone to just use an API?

HeyGen allows you to do for this in a few ways

Just skimmed this but it immediately gives up on meshes :( i actually met with a startup, backflip.ai, that is doing mesh generation and their results are very impressive. But it’s a commercial thing

Most recently: https://www.theinformation.com/articles/why-gemini-probably-...

Past: https://www.vanityfair.com/news/2018/05/uh-did-google-fake-i...

There were a few other agent demos around 2017-2020ish that also were faked. It's poisoned the well for me at this point so I just have to wait and see

I counted very quickly but 78? I learned arabic in kindergarten, im not sure what your point was. There are arabic spelling bees and an alphabet song just like english

The comment you replied to was saying LLMs trained on english cant count letters in english

If you have to create new types then why can’t this just be done by existing type systems, eg in Typescript:

Linear<T, S> = { state: S, data: T }

type CompletedTxState = ‘commit’ | ‘rollback’

Then eg for db transactions:

const commit = (t: Transaction, dbconn: DBConn): Linear<Transaction, CompletedTxState>

Looks awesome, but all of Google’s multimodal assistant demos of recent past have been faked :/ will believe it when I can use it

This is a really cool paper, reminds me of the simple exercise Karpathy goes through in his NN vid series with a bigram predictor. Looks like in practice there’s still some grounding issues when attempting to use them for instruction-tuned applications, but clever direction to explore!

I thought it was so over the top it transcended condescension and was firmly comedic. Gilfoyle from HBO’s Silicon Valley could’ve written that.

Well, when one group disenfranchises you via military conquest without such a Noble Distinction, and you cannot defeat them in a Noble Battle, you expand the available choices. And it seems to have worked out for most independence movements.

Yes, generally people side with their family and loved ones over principles. My fathers family had their house aerial bombed by the french, so a few tram bombs just sounds par for the course of an independence uprising. But your main point is correct - violence is always ugly, and we have to be careful when rationalizing it to achieve pragmatic goals.

This seems really low. If we could throw 3.4b at a new car company and get the awesome suspension tech and price points of the BYD cars then that’d be huge. If every country could do this domestically, even better!

That evaluation would be an amazing addendum or engineering blog post! I know it’s not as sexy as announcing a product, but from an engineering perspective the process matters as much as the outcome :)

Did you try clickhouse? What were its weak points?

Im currently waiting on the motors for this, but my bambu p1s printed out the parts with minimal stringing in like 90 mins. Really to try it out for cooking experiments

I see, I agree with you. But I would imagine the useful metric to be “error rate below X GB memory”. We really just need memory and/or compute reported when these evaluations are performed to compile that. People do it for training reports since compute and memory is implicit based on training time (since people saturate it and report what hardware they’re using). But for inference no such details :\

Many applications dont want to host inference on the cloud and would ideally run things locally. Hardware constraints is clearly important.

Id actually say its the most important metric for most open models now, since the price per performance of closed cloud models is so competitive with open cloud models, so edge inference that is competitive is a clear value add

What does borderline dishonest mean? I only read the abstract and it seems like such an obvious point I dont see how its contentious

you're right, but people predict the future all the time, if there's a pre-established model for that prediction. I just threw a ball in the air and predicted how long it'd stay there </akshually>

so I think this comes down to how the predictions were arrived. One way to do this is to ask individuals who both have similarly high predictive scores and bet on the same types of events to explain some of their past predictions, and if their methods are similar then you've learned a new predictive tool.

I think youre kind of right, its not really meaningful to know “im right about things that happen 5% of the time!”. But if a lot of things you bet on structurally happen to be low probability and of the same type (eg you bet on one type of catastrophic event like pandemic, crop disease, or war etc) then its useful

This is very cool, the camera over world example really highlights the difference

So the main idea is that teachers can use examples with PII/user data from gmail etc, but distill task-specific capabilities to students that don’t contain PII/user data. Seems kinda useful for companies that want to utilize private information safely.

Edit: misunderstood the paper on first skim, they don’t actually want smaller students, they might actually want bigger ones. So this is more a privacy thing

Well, GPU failure modes (from what Ive heard from ML infra people) are often subtle, eg incorrect multiplication results. So its not as simple as the usual ‘treat resources like cattle not pets’ because tou dont know which cows have mad cow disease before committing to an expensive training run

I think Google denied putting the play store on the quest (Boz from Meta has said this). Thats probably a reason this potential partnership was unlikely

the actual conversational flow is awesome. 800ms is only a little worse than internet audio latency (commonly 300-500ms on services like Discord or even in-game audio for things like Valorant)! Also cool that you can bring your own LLM and audio provider. awesome product!

AI comp is starting to form a 4th mode in the popular “trimodal” comp (https://blog.pragmaticengineer.com/software-engineering-sala...)

Have friends (not at OpenAI) that have been offered $1m+ yearly (these are not academic superstars either, solid staff engineers) to leave their startups and join Big Tech orgs.

Good for them and good for the people of OpenAI, their revenue growth has proven me wrong and is really exciting for the future of productivity.

Cool! I haven’t tried SigLIP out yet but it seems to be the new hotness over CLIP… I just dont have a good project idea yet

I was hoping that the Vision Pro would have camera access, so that developers could actually build interesting mixed reality/augmented reality apps. As soon as I saw that Apple locked down the cameras just like the Meta quest 3 I knew I'd be $3500 richer. Maybe one day...

it's always better to have a product a few people love than something a lot of people like.

Instagram and Whatsapp? Even Quest as a platform has tens of millions. All of these are still up and to the right in terms of user numbers and revenue