Same with Astra as well. I have been trying to describe this for a lack of better wording. But for so many things we need to measuring from 100%, and not measuring from 0% where 93% and 100% are both high number and similar.
If 100% is 0% error and 90% is 10% error. From this view, 90% is 10× worse than 99% in terms of remaining error or failure rate. Going from 99% to 99.9% is another 10× improvement, and so on. The differences are orders of magnitude, not linear percentage points.
I am not entirely sure I will accept the cost cutting measure of both Sol And Astra. And if people say this is not considered nerfing I think I will disagree.
Different products and services will have different expectations. But If a product has already reached 100% then all subsequent efficiency improvement should reach 100% if not 99.9% while being faster or cheaper.
A bigger potential issue is that 100% means the metric was saturated--and we don't know by how much. Sol 6.1 could have had 1.01x the capabilities it needed to hit 100%, or 10x.
Not really, because 99.9% vs. 90% and 49.9% vs. 40% are the same difference in percentage points but not at all in what they are describing (which is closer to log of odds ratio).
But they’re also not just talking about percentage (except to the extent that you can express any ratio as a percentage) since 99.9% to 90% is a decrease of both 9.9 percentage points and approximately 9.9 percent, which doesn’t convey that it’s a 100× increase in error rate.
I had to check wiki, IE 11 came out 13 years ago! Internet Explorer 11 support ended on June 15, 2022, and it was permanently disabled on February 14, 2023. ( must have missed the memo or I have forgotten about it )
W3Counter shows Around 0.07% of market share.
I wonder why IE11 is still being used or even a requirement? Especially in today's world where security could be a very serious issue.
IE 11 isn't fully out of support for a few more years yet. There are so many enterprises using it with so many custom tools built just for the browser. Most is tight lipped under NDAs but some of the details and remaining usage figures are staggering.
As someone who dont follows closely. How do I know at what point were the other GPT model tested?
Let say they just nerfed GPT 6.1 a few weeks ago. Now they are using that as a reference score against 6.2? Or were those results tested when 6.1 relesed. What about 6.0?
And that is the same with Claude. How do we know they are not cherry picking the date and test on these release when they are constantly "enhancing" their model.
We also know some of these benchmarks are largely irrelevant once they were known ahead of time and become a game of benchmaxxxing. It just follows Goodhart's Law "When a measure becomes a target, it ceases to be a good measure."
Anecdotally, GPT 6.1 Sol has turned to shit over the past month but I don’t doubt the benchmarks report no change. I wouldn’t nerf benchmark runner’s accounts either.
I've been testing sol, and opus, almost weekly as new models are released. I haven't seen any evidence of the nerfing conspiracy theories. https://slopcop.com/power-ranking
Edit: But it does look like you get about 20% less tokens for your subscription with OpenAI than you did until a couple months ago.
No one has provided you a clear answer on why your contrived scenario of an individual bringing in 12 GPUs to Singapore is not relevant at all to what NVIDIA is being accused of doing?
That is great, more than happy to hear the HN community doesn't engage with sand-bagging at this level
No one has even answer how Nvidia is doing wrong. Apart from quote "I am sure".
Except for the rest of us keep pointing it is not wrong. And provided evidence Megaspeed is not blocked, and how it ends up in China ( or not ) is not Nvidia's fault. And renting is not against sanction.
All all we have are less than 1 year account keep pointing out the same thing. I sometimes wonder if it is smear campaign.
Sir Humphrey Appleby: "No, Minister. Either you sell arms or you don't. If you sell them, they will inevitably end up with people who have the cash to buy them."
James Hacker: "But not terrorists!"
Sir Humphrey Appleby: "I suppose we could put some sort of Government Health Warning on the rifle butts: 'This gun can seriously damage your health.'"
When I first watched it, I thought "Haha, a biting satire that greatly exaggerates flaws and failings for great comic effect!" Now, when I watch it, I think, "Oh, what a utopian dream! If only our government was this well-meaning and competent."
I know this sounds stupid, for the past 5 - 6 years I have increasingly wish I go back to tungsten light bulb. Or wish we could somehow make tungsten a lot more efficient.
At this point I am not even sure why should we continue to read anything from mainstream media. Other than some sort of civil duty to continue correct them or bring it context / communities notes.
I am also wondering why Datecenter and AI is now becoming the new Anti-Christ / Nuclear Reactor.
I think this perspective lacks nuance. If you are so pro-data center you don’t understand why it there is so much fear around them, start there. You don’t need to change your mind, but it is important to understand other perspectives. Otherwise you are going to stay in your own bubble and continue to avoid news that aligns with your perspective.
I am neither pro or against anything. But nothing in the Anti Data Centre side has provided anything concrete. Are there downsides? Yes. But at this point even without any down side people will still be against it.
Honestly, why would people be "for" datacenters in their neighborhood?
Heavy industry gets/got a lot of pushback, and that brings both local jobs and outputs that are essential to many.
Datacenters, on the other hand, bring basically no local jobs once construction is finished, and--worse even--their output is perceived to compete with the job people already have to some degree.
If 100% is 0% error and 90% is 10% error. From this view, 90% is 10× worse than 99% in terms of remaining error or failure rate. Going from 99% to 99.9% is another 10× improvement, and so on. The differences are orders of magnitude, not linear percentage points.
I am not entirely sure I will accept the cost cutting measure of both Sol And Astra. And if people say this is not considered nerfing I think I will disagree.
Different products and services will have different expectations. But If a product has already reached 100% then all subsequent efficiency improvement should reach 100% if not 99.9% while being faster or cheaper.
reply