HN user
I say don't muddy the water with the public panic over "will it won't it" bubble burst predictions.
The effective demand for Opus 4.5 is bottomless; the models will only get better.
People will always want a code model as good as we have now, let alone better.
Bun securing default status in the best coding model is a win-win-win
Cooperation with the eval admins
not wrong, per se, but it's far from the first time
depends on the hyperparams but one of the biggest benefits of a latent space is transfer between modalities
in a bunch of azure data centers?
ASU's Center for Science and Imagination has a project[0] specific to this that they launched[1] in 2014
[0]https://csi.asu.edu/category/optimism/ [1]https://csi.asu.edu/project-archive/optimism/science-fiction...
Lovely CSS and nice content chunking, but perhaps overly biased towards finding utility with LLMs as there is too little information about cautions to take.
Most egregiously, on the AI is stateless page[0], it shows relying on an llm reponse to call a program via sudo. I can't think of a system where a professional programmer could justify such a risk.
I think after "AI predicts text" it makes less sense to say "AI is just another API" than something like "AI makes APIs fuzzy" because the stochasticity makes an AI API anything but "just another API." Most people think of APIs as deterministic in function, AFAIK.
I /feel/ similarly intuition-wise. But models are crazy and what they respond to is often unpredictable. There are no lungs in an AI model but nonetheless 'take a deep breath' as a prompt has shown[0] improvement on math scores lol
Personally I strongly disapprove of the first/second person pronouns and allowing them [encouraging, even] to output 'we' when talking about humans.
[0] https://arstechnica.com/information-technology/2023/09/telli...
Possibly because it dovetails with pharma mfg and [potentially] food mfg. Could see a case made for enzymatically brewed 'meat inks' [very sorry for this term ;p] for 3d printing the next gen of lab meats.
Certain based on anything other than instinctive surmise?
The massive LLMs trained on webscale data aren't. But some are, in fact:
It isn't clear at this point that the Azure thing and the CS thing are unrelated
Largest context window (because of most compute) wins.
They've got more serious engineering heavyweights, putting a lot of collective work on fewer tasks/approaches. Microsoft is taking more of a kitchen sink approach
Pretty similar to cappy https://arxiv.org/abs/2311.06720
Gemma 2 27B is currently the best performing 'open' model [license is non-commercial].
The Unsloth team have a blog post up where they've made fine-tuning Gemma 2 require less VRAM, and also have extended the context window.
They've also updated their 'mistralified' PHI-3 models to Microsoft's June update of PHI-3 which sees some performance increases as well.
It's never been better than what MS had and if you think so you've never used the MS way as I do.
Alt+underlined letter with no menus opens opens whichever menu maps to that letter
After which any [unique, as they mostly are] underlined letter activates the press for that menu item.
There is no equivalent to the speed of this in OSX, even with third party software.
Very very happy to be proven wrong on this
No it doesn't. Alt+space bring up the window system menu to restore/maximize/close?
Alt+f, a to save as?
Negative.
If you pay attention to what I wrote vs what you wanted to read, you'd see the false equivalence is yours.
Pay-gating copyrighted content is different indeed from public-posted social media content. In the instance of the Forbes theft, Perplexity offered links to other copycats, but never the original [costly] reporting.
Never coded a GUI app but speaking of regressions:
As a heavy windows KB user, losing the ability to alt+letter all the things to /quickly/ navigate windows apps is incredibly frustrating.
I've read that Mac OS envy infiltrated the WinUI time and while some may prefer the aesthetics, on Windows 11 apps like mspaint I can no longer navigate anywhere near as quickly.
From milliseconds to multiple seconds.
I'm incensed at this change personally, and I feel totally confused by it from the perspective of MS. In the era of AI and automation, slower screen draw times will make a significant impact to performance and energy use. I hope someone like Mark Russinovich, Kevin Scott, or Satya Nadella will notice and change course.
Apply that idea to the journalism perplexity stole, which obviously has value. Forbes worked for months to create a story only to watch an IP launderer garner more financial reward.
You mean like this?
Have you seen Zamba by chance?
May be some here https://github.com/iipc/awesome-web-archiving/blob/main/READ...
Your enthusiasm is great! People don't want to quash your enthusiasm, and I'm in the same boat.
But while enthusiasm is great, delusion is not. Since you're striving to be a founder and not a hobbyist, you have to be realistic about what you're trying to build.
What you're describing is fundamentally not possible to provide assurances on without some kind of legititmate AGI, which you lack the resources to build yourself.
Many better resourced companies are trying to provide grounded, factually accurate information, so it just seems like an area of effort far too broad to ever succeed in.
I would suggest a pivot into demonstrating legitimacy in a very narrow niche before attempting to be a genralist know-it-all. Providing fine-tuning as a service to a point of assured factual grounding is itself a hard enough open challenge in AI.
Might not be a 'coordinated attack' so much as the consequence of a referral[0] program in the age of AI
[0] https://docs.google.com/forms/d/e/1FAIpQLSe52_7L-JqY6OqhL0FJ...
You really should know better than to interrogate an LLM about itself. They do not have self-awareness and will readily hallucinate.
"Meta also announced a partnership with Google to include its real-time search results in the assistant's responses, supplementing an existing arrangement with Microsoft's Bing search engine."
from
https://www.reuters.com/technology/meta-releases-early-versi...
Have you looked at AICI by Microsoft yet?
Will it cost any less? [Apart from not tipping] As inconsistent as the "to the door" service can be, for some of the enormous apartment complexes here grabbing from the car is going to be a major barrier to this being preferred.
We're onboarding to Cloudflare MagicWan and want to use them for logging, which they do to 's3-compliant' buckets.... on Google or Amazon.
I was pretty surprised at the lack of dogfooding, wondered if it's an oversight, on somebody's Gantt, or just not something R2 can handle for some reason.