>> MoE model with 52 billion activated parameters means its more comparable to a (dense) 70b model and not a dense 405b model
Only when talking about how fast it can produce output. From a capability point of view it makes sense to compare the larger number of parameters. I suppose there's also a "total storage" comparison too, since didn't they say this is 8bit model weights, where llama is 16bit?
For decode steps it depends on the number of inputs you run at a time. If your batch size is 1 then it runs in line with active params, then as you get to like batch size 8 it runs in line with all params, then as you increase to 128ish it runs like the active params again.
For the context encode it’s always close to as fast as a model with a similar number of active params.
For running on your own the issue is going to be fitting all the params on your gpu. If you’re loading off disk anyways this will be faster but if this forces you to put stuff on disk it will be much slower.
There's always a bit of incredulity on peoples face when I tell them I enjoy reading science fiction books.
I understand why they would react that way.
Studies consistently show that young adult males, typically between the ages of 18 and 34, are the most avid readers of science fiction. Here are some statistics to support this claim:
1. A 2019 survey by the Pew Research Center found that among adults aged 18-29, 47% of men reported reading science fiction, compared to 34% of women in the same age group.
2. A 2015 study by the publishing industry website, Goodreads, analyzed the reading habits of over 40,000 users and found that 62% of male users aged 18-24 listed science fiction as one of their favorite genres, compared to 45% of female users in the same age group.
3. A 2013 survey by the market research firm, Nielsen, reported that 55% of males aged 18-34 read science fiction compared to 41% of females in the same age group.
All this statistics goes to prove why most people are surprised when I tell them I enjoy reading Sci-Fi, this statistics is what also pushed me to create this site, a website where both genders can enjoy reading Sci-Fi tailored to them whether they are twenty something or 60 something.
Q: does anyone think it possible to rival them in a decade with enough funding?
> go build your own "real competition" if you think its that easy.
Another dismissive comment. What inspires these? The Q proposes a significant timetable and suggests adequate funding. I'd agree it's open ended but it sounds like the OP is inviting education on the topic.
My God, this is truly beautiful. If this were the blurb of an actual novel, I would unhesitatingly drop EVERYTHING to read it, and I say this as someone who reads an average of 1 novel every 3 days.
If LLMs can replace 90% of your queries, then you have very different search patterns from me. When I search on Kagi, much of the time I’m looking for the website of a business, a public figure’s social media page, a restaurant’s hours of operation, a software library’s official documentation, etc.
LLMs have been very useful, but regular search is still a big part of everyday life for me.
Sure we now have options, but before LLMs, most queries relied solely on search engines, often leading to sifting through multiple paragraphs on websites to find answers — a barrier for most users.
Today, LLMs excel in providing concise responses, addressing simple, curious questions like, "Do all bees live in colonies?"
Media. They own YouTube and are backing away from alt-right mainstreaming 'cos the money went out of it, turning instead to direct monetization. If their previous path was still profitable they would still be doing it and we would hear nothing of adblock wars, as that highlights one of their problems for which they are directly paid. I don't think there's anything else in that category to speak of.
They have the power to crosslink this monetization to the success of people using the platform by just making it part of the algorithm. Pay to play (your videos to any audience) and in so doing, normalize that we pay for this stuff like it's cable. Their economies of scale mean their break-even point is way, way lower than it would be for anybody else.
Maybe it would be unethical to stifle people on your platform if they're not normalizing the consumption of your platform like it's a for-pay service instead of some magical free thing that comes from nowhere, but it really never was a magical free thing, and Google's ability to platform or stifle people is the most powerful force they could possibly call upon.
It's just that they're turning to an actual market now, rather than maintaining the pretense that it's all free and instead getting paid by… what, Russia? Prager U? What has changed in recent years to imply that getting paid by a booming, capitalist market might be more profitable than trying to get paid by malefactors?
I'm pretty sure Google owns media now. That's what they won. We've seen worse than what we're currently seeing as they try to fight adblock and get paid as a service. Remember what happened when everyone thought it was infinite media for free. Someone was still finding a motive to pay 'em, but hidden motives are a lot more worrying.
The posts on the site are the original posts. To seed the site so it wouldn't look like a wasteland, I emailed a few high value HN posters who had email addresses shown on their HN profiles and begged them to post a tip.
A few of them graciously obliged, and some of those probably told a colleague/acquaintance to post a tip as well.
It's why I love HN, there's more of a sense of community here.
reply