Depends on the workload. H100 will never have the network performance of Vera Rubin. There's also token per watt, newer systems will beat the older systems.
It's not clear how much of the latest chips have even made it on-line yet.
The claims of many GW of installed training/inference have come under scrutiny lately. The first VeraRubins aren't even there yet, so it's all GB300 NVL72s as the peak performers and probably <<1GW of those so far. Even xAI Colossus is mostly H200s and B200s.
Electricity costs are also a huge differentiator. When drawing 100kW the difference between >50cents and <10cents per kWh is pretty big! One is almost $0.5M and the other is less than $100k.
Nerd sniping the hackernews/twitter crowd with "large scale transformer-based language model alternatives."
People don't want to believe something as unsatisfying as "Scaling up LLMs" can yield something as profound as AGI/be useful, and just hope that literally anything else can take their mindshare away, and this just happens to be the new rage. Along with clearly-not-frontier-level open source models, non-transformer based architectures, etc.
Beats me also - this feels unreliable, extremely niche, and over-hyped. I don't trust LLMs even when they explain their reasoning; the idea of trusting a black-box classifier like this seems insane.
In my company, and I think in most companies that are using AI at all, one of the first ways it got integrated is as a classifier, to tag orders based on feeding all their data into a prompt and asking for a structured output.
I think demand for tools that are more tailored for this type of integration is high. I don't really understand why Jev is supposed to get my company's decisions right more than an LLM, but regardless of the tech I think people are just excited about the possibility of iterating faster, more explainability, higher-level tools that are specifically created to help hone classifiers etc.
I think the point of Jev is to thread the needle of the gap between non-LLM classifiers and LLMs.
Classifiers like classical NNs require:
- annotated data, potentially a lot of it
- training
- inference
#2 and #3 aren’t a big deal if you have an ML engineer, but #1 will always be a potential headache no matter who you are. The tradeoff is that they could be quite fast, cheap, and you can get probabilities, not just classes.
With LLMs you get:
- zero shot classification (no dataset or training required)
- potentially can use third party model providers like OpenAI off the shelf. Don’t even need to host your own model.
The downside to LLMs is that they are comparatively slow and expensive to traditional classifiers. Historically they also were prone to hallucination or malformed responses, though not as much these days. You also can technically get log-probs back, but these aren’t equivalent to the classifier probabilities.
Jev gets you the zero-shot, zero-infra benefits of LLMs, while being closer to the speed and cost of traditional ML classifiers, as well as both classification and probability responses.
Yeah but they weren't that great, you couldn't ask for arbitrary classifications after the model was trained. You are underestimating what they've done here, even if it does seem a little overhyped.
Google's Android customers are phone OEMs. The major ones seem to like how things are going. Until Samsung and whomever start the OpenHandset Foundation or some such and fork Android, there's never going to be the Mariadb of Android.
Depends on what you qualify as 'standard of living.' There's no amount of money I would accept to live in many parts of the world. To have the same size property and home I have today in France or Germany would be likely 10's of millions of Euros. My property is worth well less than $1M.
And we know this was an outright lie, because the clinical trials did not include contracting the virus as an endpoint. In fact, in Pfizer's data, they only used PCR tests on symptomatic people, rather than periodic testing.
So to say the clinical trials reflected absence of infection is a deliberate lie. Either they knew what the data said and lied, or they didn't know what the data said and stated they did, which is a lie.
Pfizer didn't lie. If you go back to their press releases they were careful to distinguish SARS-CoV-2 from COVID-19, the result being that they said you wouldn't get sick while saying nothing of infection/spread. It was the media and politicians that started that.
Debatable. For one, many people believed the stats were cooked. Hospitals had financial incentives to claim corona cases. They routinely didn't test people for corona if they were vaccinated during the worst part of the outbreak.
Even Pfizer's own trial data submitted to the FDA showed an all-cause mortality higher in the control group than the product group. Of course, they explain all the deaths away.
You're aware that the J&J was pulled from the market due to cardiac issues, right? It's not a theory that actual people were actually harmed by the products, the only question is risk/reward.
More cardiac issues than the other vaccine options but far less than the virus. If it were the only option it would still be on the market. It was only pulled because better options showed up.
Unfortunately, this isn't a claim that can be made. We don't know how many people got the virus, or how many times. And IIRC, the cardiac issues of the virus were mostly in older demographics, the J&J was affecting young a healthy people.
> the cardiac issues of the virus were mostly in older demographics, the J&J was affecting young a healthy people
I have a friend who got Type 1 diabetes after Covid attacked his pancreas. (And he got infected in March 2020, so this isn't someone choosing to be diseased getting their just desserts.)
> the J&J was pulled from the market due to cardiac issues, right?
Yes. I'm saying now that we've had time to examine those cases and look at the data, how many people are clinically agreed to have actually suffered long-term harms? (I don't believe the myocarditis was a long-term effect.)
reply