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

Nvidia is among the best positioned if open models win.

Most of nvidia's revenue comes from data center scale out. They will loose their ass if local models win.

open models don’t imply local models… it just allows more people to run them, most likely with nvidia GPUs

Any model ran takes compute. Everybody, datacenters and local need compute.

And I don't see a future where datacenters lose much to open models. I've ran local server rooms before clouds were a thing. Cooling and power is a very expensive pain in the ass. You're not dropping a rack of local models in most buildings without breaking the local power grid.


What about when the open models get to the point that you don’t need these fancy GPUs? It seems inevitable this will happen over time, we see cracks right now within even just GPT w/ expert streaming and flash attention. ASICs are coming soon too.

Nvidia already have a plan in place for fast inference. They acquihire Groq last year

AWS and Azure would love to cut out OpenAI and Anthropic and sell inference on open models directly to businesses with no cut to share, just as AWS serves other open source software.

Open models are great for everyone except anthropic and OpenAI.


I also don't see a strong link between training at a competitive activity, and releasing a product. Obviously, training against real opponents is crucial.

But the world is full of half-baked products, especially software. If you show up to your first tournament and lose pathetically in the first round, no one is going to prevent you from attending the second tournament. But people will permanently write off your product if it is half-baked. Or even if your product is good but the release process is half-baked.


At the same time lots of half-baked products are successful and generate real value for the users and the creators.

And some people win the first tournament they attend. Again, flubbing your first tournament is much less persistent than flubbing a product release. Significantly different risk profiles.

Bad software, as in stateless programs, doesn't actually matter and never did. They can be replaced trivially.

The problem is the real world isn't made of stateless programs, but lots of important data in bespoke formats/schemas, and if you change the shitty software that interacts with the important data, in the wrong way, you can lose everything.


OAI/Anthropic shareholder? Speed and intelligence are not "every way". Cost is essential. Hence the Pareto boundary illustrated in TFA.

It literally cannot complete tasks that Luna can do easily. It doesn't matter how cheap it is.

It actually can. I have been using it regularly for past 4 days. It is on sol-low level. I deliberately tested it on a moderately complex task. glm-5.3-flash one-shotted it correctly. Luna max couldn't achieve parity even after 3 total attempts.

It was creating java bindings for this project: https://github.com/jeffhajewski/latticedb

And here is the binding one-shotted by glm-5.3-flash: https://github.com/jeffhajewski/latticedb/pull/5


Are you implying that getting $MEGASaaS to implement a feature for you is faster than doing it in-house? Because that is absolutely not the norm.

Both of you are right. There is demand for tailored (fine-tuned) models; almost every enterprise would theoretically benefit from them.

But there are also a lot of prerequisites, namely does the enterprise have its sh*t together on a technical level. Does it have the processes and data pipelines available to train and benefit from these models? Probably not!

Applied ML is at the crown of a tech pyramid whereas most enterprises are still struggling at ground level. Being able to build from be ground is likely a safer skillset than only knowing how to work at the (non-existent) apex.


LLMs have only been really good for a couple years. You can bet that the remaining AAA studios are working hard on figuring this out in current unreleased products.


One challenge I see is that to work well and stand above 'easy' implementations the game would need to provide a rich context for the AI to exist in and interact with. Skyrim is a richly detailed world with an engine that allows that level of interaction, especially as it builds upon earlier elder scrolls games. Then there's the factor of whether commercial studios with the capacity to make that world in the first place (which has been a major draw for players) will invest further budget into AI to add more value, and if they have options to see a return on that investment via the main purchase price or additional costs.


just offer a separate subscription to host the companion model. I feel dirty typing this.


There is, it's called VictoriaMetrics/Logs/Traces.

https://victoriametrics.com/


We use victoriametrics, but I believe that's just the collector side of it cuz we also query into it with grafana.

Datadog isn't just a collector, but the whole querying UI as well, right?


Yes, Grafana seems to be the best free querying frontend across multiple data sources right now. But each Victoria product has its own built-in query GUI as well.


Would you like to share your success story using this software?

I believe I first deployed VictoriaMetrics almost 5 years ago and have had zero regrets. Was also evaluating Thanos and InfluxDB at the time.

The lead dev is super passionate about straightforward software that just works, and it shows.

Later my coworker deployed VictoriaLogs in about 15 minutes, added it to our syslog targets. Night and day compared to Elastic. Use both of them probably every day through Grafana.


Assuming they (and militaries broadly) do this +more, like actually using vision models trained on billions of geolocated landscape photos.


Installing & maintaining a Postgres extension is vastly, vastly simpler than running Elasticsearch and Kafka. Like, how could you even compare these things if you know what you are talking about?

Or maybe you are looking at it from "just swipe your credit card at AWS" perspective, in which case "just use Postgres" articles are for a different audience.


Running a small ES with dual save alongside your relational store is trivial. Kafka I will grant you is a lot to operate, but I would not choose Kafka unless I have a Kafka-shaped problem, and if it looks like that I would never, ever choose Postgres. Equating Kafka with “I need a queue” is a laughable comparison.


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

Search: