Hacker News new | past | comments | ask | show | jobs | submit | nycdatasci's comments login

NVDA can't really "churn out more CUDAs" because CUDA is a software platform/framework, not a physical product.

Apple lets their design team make all the critical product decisions, while Microsoft delegates the same responsibility to marketing.

Don't think that's true because increasingly often their marketing is nothing like the actual product. Doesn't feel like any of them are working together internally.

Microsoft still hasn’t figured out desktop search. Of course o365 Copilot is a disaster.

Amen

Models will eventually be interface agnostic and they will cover all interfaces that are commonly used by individuals and organizations. It won't matter whether you have a nicely documented public API, a traditional website, or a phone interface to customer support.

Initial investment in a data center in TX, but planning to spend up to $500B over four years!

Stargate is an interesting choice of name. A prior project by the same name investigated the potential for psychic phenomena in military and domestic intelligence applications: https://en.wikipedia.org/wiki/Stargate_Project

Slightly off topic, but also curious why Cloudflare doesn't put more effort into policing content of Pages, which are frequently used by bad actors. https://www.bleepingcomputer.com/news/security/cloudflares-d...

Examples: https://pending-revew.pages.dev/ https://r2-cmq.pages.dev/ https://ampgoat-ligaciputra.pages.dev/


Metaphorically policing content is expensive and exposes you to politics. Everyone is trying to get out of doing it (see: recent Meta announcement).

If you are the literal police, they will do something.


Lots of complaints mentioned here. If you have a legitimate need for a product like Tasks that is more fully baked, I’d encourage you to check out lindy.ai (no affiliation). I’ve been using it to send periodic email updates on specific topics and it works flawlessly.


I've done head-to-head tests vs. Claude 3.5 Sonnet on a variety of coding tasks and basic reasoning questions over the last week. It's not worth $200. Claude is better on most tasks.

For example, I pasted my HN profile and asked it to extract my email address in JSON format ({"email": ""}). I would expect any HN reader to be able to do this in seconds using an online ROT13 calculator. Certainly a "phd-level reasoning model" should be able to do this correctly. Claude outputs the correct answer in seconds. o1 Pro thought for two minutes and eventually output an email address that was invalid.

I'm a little surprised that there isn't more discussion about this on HN, as it seems highly relevant to the recent pivot from training-compute to inference-compute.


Definitely not luck. Bill runs with the same crowd that wrote the bible on advantage play (James Grosjean, etc). They are exceptionally talented.


Bill originally worked with a team of people doing this in HK (that later broke up, some of the people involved were...characters). There was a whole group of people doing this in the 90s (some of these teams also did Japan). I believe he did stuff in casinos but parimutal obviously had a lot more scale.


Exactly right. Impossible for individuals to operate at this scale. Bill led a large operation. Similarly, advantage play frequently requires teams to execute as you are often hoping to land a specific seat with a specific dealer.


Consider applying for YC's Spring batch! Applications are open till Feb 11.

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

Search: