It very much looks like a side effect of this new architecture. In my experience, text looks much better in recent DALL-E images (so what ChatGPT was using before), but it is still noticeably mangled when printing more than a few letters. This model update seems to improve text rendering by a lot, at least as long as the content is clearly specified.
Yeah who wouldn't love a dip in the sulphur pool. But back to the question, why can't such a model recognize letters as such? It cannot be trained to pay special attention to characters? How come it can print an anatomically correct eye but not differentiate between P and Z?
True, that would be preferable, but alas Samsung is bent on making their products as big of a pain in the arse as possible.
At least with my Samsung soundbar, the remote can change the volume, the subwoofer volume and change between modes (standard, surround, game). But if I want to enable night mode, I have to use the SmartThings app. There's no way to enable it using the remote. What's worse, the app often hangs when connecting to the soundbar, requiring me to force stop and restart it. So sometimes toggling a feature that should be a single button on the remote takes me over a minute.
Samsung is right next to HP on my list of brands I will never ever buy in my entire life.
If 1.2 PB is a problem, then why don't they just specify a bandwidth limit of say 1 PB? They specifically say "unlimited bandwidth", so yes, what they are doing _is_ scummy because there is a very obvious incongruity between what they claim and what they actually offer.
I imagine because people will immediately push up against the limit and no further. It’s much easier to detect these excessive users if their bandwidth naturally keeps growing.
This is just stratified pricing. If you're egressing 1.2PB ($50k+ worth of bw on AWS) there's a likelihood that you're earning a fair bit and an enterprise contract will be worth it to you when it comes to support. On the other hand if you're egressing 1.2PB to serve some open model weights that you don't charge for, CF would prefer to leave you to it and enable you to serve.
The fact that this has just disappeared from the front page for me, just like the previous post (https://news.ycombinator.com/item?id=41651548), somehow leaves a bitter taste in my mouth.
Look at the URL. It’s because the original WSJ title was “OpenAI Chief Technology Officer Resigns,” which was a dupe of yesterday’s discussions. WSJ changed the title yesterday evening.
How do you explain all of the constant unflagged criticism of OpenAI and Sam Altman throughout nearly every OpenAI thread? I mean, look around at all of the comments here...
Same thing in Germany, with WetterOnline ruminating data from DWD (the German meteorological service) and then suing them when they offered their DWD WarnWetter app for free. Unfortunately, they were successful: https://www.heise.de/news/BGH-Urteil-Staatlicher-Wetterdiens... (Couldn't find an English source, sorry.)
I'm continually amazed by the (lack of) quality of their laptops and the scenario described really reinforced my opinion of HP.
At work, I've used a ProBook, a ZBook, and two EliteBooks, all of which had major issues. Sleep mode never worked on any of them (immediately turning back on again with powercfg /lastwake showing no reason), and my current EliteBook frequently shuts off without warning and then won't turn on for five minutes. The ProBook and one EliteBook failed randomly and needed to be replaced.
The ZBook's workstation CPU overheated even at light usage, making it unbearably slow. Despite IT saying nothing could (and should) be done, I disassembled it and found it was missing thermal paste, or what little there was had hardened into a brittle, useless mess. Reapplying thermal paste about tripled the Cinebench score.
Given all this ridiculousness, I can't imagine how much worse their consumer laptops must be. It's baffling how anyone but the most naive non-tech people still buys from them.
My manager's company-issued HP EliteBook is actually the third device he's had during the last three years, as the previous units failed without warning. Nowadays if I don't see him on the standup I assume it's this problem again.
Kind of reminds me of the darkest days of MBPs and their failing keyboards.
My work laptop is doing ok so far with only minor annoyances like needing to reconnect peripherals after waking up the device, but that's it.
Meanwhile I cooked the screen in my Asus personal machine because it assumed that sleep = 100% CPU. Thermal paste was of course cooked as well, so I had to replace it.
All in all I'm glad that Framework expanded into my country recently. It's expensive for what it offers, but half the reason I'll be ordering one is spite.
They claim that the new models "are significantly less likely to refuse to answer prompts that border on the system’s guardrails than previous generations of models", looks like about a third of "incorrect refusals" compared to Claude 2.1. Given that Claude 2 was completely useless because of this, this still feels like a big limitation.
The guard rails on the models make the llm-market a complete train wreck. Wish we could just collectively grow up and accept that if a computer says something bad that doesn't have any negative real world impact - unless we let it - just like literally any other tool.
They're not there to protect the user, they're they're to protect the brand of the provider. A bot that spits out evil shit easily screenshotted with the company's brand right there, isn't really great for growth or the company's brand both.
True and this is also the reason why open source models are commonly uncensored.
It's frustrating though because these companies have the resources to do amazing things, but it's been shown that censoring an LLM can dumb it down in general, beyond what it was originally censored for.
Also, this of course. It's just a cheap bandaid to prevent the most egregious mistakes and embarrasing screenshots.
I don't disagree but on the other hand, I never run into problems with the language model being censored because I am not asking it to write bad words just so I can post online that it can't write bad words.
Hm, I don't buy this. The statistics shown in the blog post revealing the new Claude models (this submission) show a significant tendency to refuse to answer benign questions.
Just the fact that there's a x% risk it doesn't answer complicates any use case unnecessarily.
I'd prefer if the bots weren't antrophomized at all, no more "I'm your chatbot assistant". That's also just a marketing gimmick. It's much easier to assume something is intelligent if it has a personality.
Imagine if the models weren't even framed as AI at all. What if they were framed as 'flexi-search' a modern search engine that predicts content it hasn't yet indexed.
Yeah I spent a lot of time with Claude 2 and if I hadn’t heard online that it’s “censored,” I wouldn’t have even known. It’s given me lots of useful answers in close to natural human language.
>It tells me at a glance whether some area is a forest or just a field. Everything's just the same color in Google Maps.
Thank you. Everyone is talking about the color of the water or the roads, but this doesn't even bother me. But having all vegetation of any kind in the same color is an absolutely mind-bogglingly stupid decision. I can't be the only one who uses Maps to check the extents of a forest, for example. Or who uses shapes of different land uses to quickly orient oneself on a map.
Yeah, it's quite clear the designers have never lived outside lived of a car sewer. Every justification they've given thus far circle back to contrast for road features and while driving. It's peak car brain.
Car sewer? Car brain? Wow, didn't know we had zealotry and disparagement over _choice of transport_ of all things, generally a matter people have little choice in; it's determined by where you live and what sort of life you have.
It's not the network that's the bottleneck, but the Javascript-bloated mess of the frontend.
edit: That isn't to say that Jira and Confluence are the only offenders. They aren't even the worst. Slack/Teams/Rocket.Chat are all just as bad and make me consider switching to a career without computers on a regular basis.
That’s kind of my point; all the tech you list here doesn’t have horrible lag, and aren’t generally even close to so bad they’d make someone want to switch careers.
So if it’s not them…
However, when giving a prompt that requires the model to come up with the text itself, it still seems to struggle a bit, as can be seen in this hilarious example from the post: https://images.ctfassets.net/kftzwdyauwt9/21nVyfD2KFeriJXUNL...