he aged

I don’t want to be forever young, but I’d love to feel like I’m in my 20s until I’m 100.

The critical thing to remember about LLMs is that they are probabilistic in nature. They don’t know facts, they don’t reason, they don’t evaluate. All they do is take your input string, split that string into tokens that are about 3-4 characters long, and then go back into their vast, vast, pretrained database and say “I have this series of tokens. In the past when similar sets of tokens were given, what were the tokens that were most likely to be associated with them?” It will then construct the output string one token at-a-time (more sophisticated models can do multiple tokens at once so that words, phrases and sentences might hang together better) until the output is complete (the probability of the next token being relevant drops below some threshold value) or your output limit is reached.

I love how all of the characters are scowling and have their game faces on… and then there’s Kirby, who’s like “Hi there! I’m gonna eat you and extract your power!”

It would be ideal If the big activitypub platform stacks like mastodon, Lemmy, etc could agree on some standard like a federated OIDC or DID approach for all authx/authn functions. then fediverse users could get cross-platform and even cross-instance logins “for free”

Of the changes made last week to the license, this one stands out:

10. None of the Work may be used in any form as part, or whole, of an integration, plugin or app that integrates with Atlassian’s Confluence or Jira products.

That is a weird carve-out, so I’d guess the license revision (and technically the reason it’s no longer open source) somehow has to do with Atlassian or their plugin marketplace?

Feel like the (totally impractical) fediverse end-game would be for each individual to have their own activitypub service, and federation happening on a person-by-person basis. So you retain some control over anything you publish, and your history is yours to keep.

As others have said, changing UPS batteries is required maintenance, and I agree 18-24 months is the typical service life for even high-end UPSs. However, you may want to look into LiFePO4 based UPSs, which can handle many more charge-discharge cycles and often have 5-year warranties. More expensive and potentially not as recyclable as lead acid batteries, but maybe appropriate for your use case.

Yeah I wish they’d just transcribe their youtube videos and make blog-style posts with some of the screenshots. I’m almost never going to watch a 20-minute video, it’s not like they’d be losing a viewer.

Good suggestion! Though sometimes they’re a little too technical for me.

With AnandTech gone and TomsHardware aside, what’s another great “deep cuts” tech news and reviews site?

Putting aside why you’d want to do this, it’d be pretty easy, actually. You’d still use a big model like GPT4 or Claude as your “base” but you would do two things:

• Give it a knowledge base using your conversatons. You can manually vectorize them into a key-value database like Pinecone and build yourself an agent using a toolchain like Langchain, or just use a service (OpenAI Agents lets you upload data from your browser)
• Have one of the big LLMs (with a large context size) ingest all of those conversations and build out a prompt that describes “you”

you would then

• Feed that generated prompt (with your own edits, of course) back into either your custom Langchain agent or OpenAI Agent

This is a good, short read. For those who are unfamiliar with the AGPL license that the author proposes we all start using, the main difference (and I am not a lawyer) is that under the AGPL, the source code including any modifications must also be made available to all users interacting with the software over a network. This prevents companies from making proprietary versions of AGPL software that are only accessible as a web service, which is one of the big ways that corporations are able to profit from GPL source code contributions these days.

When did we get away from saying “X - formerly known as Twitter” ? I liked seeing that gentle nudge in every headline.

New drug: 60% survival rate with no disease progression.

Current-best drug: 8%

While technically it’s not “off the charts”, this is pretty damn impressive. I’ll allow the hyperbole.

Maybe true, but even at $3500 the Vision Pro would be about the cheapest thing in the operating theater anyway.

I bought one of these when the Amazon meme first got big. It has become a family tradition to try to hide it in somebody’s drawer/under a pillow/etc and claim that it was the power of the wolves. And then wait for it to make its way back….

My family’s spread out. Ours are well-traveled wolves.

This is why AI-created content will win the day. No complicated moral or ethical quandaries to navigate. Oh, except electricity usage. And copyright issues. And diminishing the value of human art and artists. And the possibility of skynet ending humanity.

Looks like it’s back to scratching rough drawings into the dirt for me…

Wow. How do they fit so much smart into such a tiny space?

Mine’s more like an LLM - exposed to a vast quantity of technical terms that they don’t really understand, but can mash them together well enough to make coherent-sounding statements in JIRA