Welcome to Incremental Social! Learn more about this project here!
Check out lemmyverse to find more communities to join from here!

Spedwell

@Spedwell@lemmy.world

This profile is from a federated server and may be incomplete. Browse more on the original instance.

Spedwell ,

Right concept, except you're off in scale. A MULT instruction would exist in both RISC and CISC processors.

The big difference is that CISC tries to provide instructions to perform much more sophisticated subroutines. This video is a fun look at some of the most absurd ones, to give you an idea.

Spedwell ,

Huh, thanks for the heads up. Section 4 makes it look like they can close-source whenever they want.

I'm just glad FUTO is still letting Immich use the AGPL instead of this, though.

Spotify is raising the cost of Premium subscriptions, again (www.engadget.com)

Spotify is officially raising its Premium subscription rates in the US come July, following reports of the move in April. The platform is increasing its Individual plan from $11 to $12 monthly and its Duo plan from $15 to $17 monthly — the same jump as last year's $1 and $2 price hikes, respectively. However, its Family plan...

Spedwell ,

There is an episode of Tech Won't Save Us (2024-01-25) discussing how weird the podcasting play was for Spotify. There is essentially no way to monetize podcasts at scale, primarily because podcasts do not have the same degree of platform look-in as other media types.

Spotify spent the $100 million (or whatever the number was) to get Rogan exclusive, but for essentially every other podcast you can find a free RSS feed with skippable ads. Also their podcast player just outright sucks :/

Why Is There an AI Hype? | The Luddite (theluddite.org)

Companies are training LLMs on all the data that they can find, but this data is not the world, but discourse about the world. The rank-and-file developers at these companies, in their naivete, do not see that distinction....So, as these LLMs become increasingly but asymptotically fluent, tantalizingly close to accuracy but...

Spedwell ,

Errrrm... No. Don't get your philosophy from LessWrong.

Here's the part of the LessWrong page that cites Simulacra and Simulation:

Like “agent”, “simulation” is a generic term referring to a deep and inevitable idea: that what we think of as the real can be run virtually on machines, “produced from miniaturized units, from matrices, memory banks and command models - and with these it can be reproduced an indefinite number of times.”

This last quote does indeed come from Simulacra (you can find it in the third paragraph here), but it appears to have been quoted solely because when paired with the definition of simulation put forward by the article:

A simulation is the imitation of the operation of a real-world process or system over time.

it appears that Baudrillard supports the idea that a computer can just simulate any goddamn thing we want it to.

If you are familiar with the actual arguments Baudrillard makes, or simply read the context around that quote, it is obvious that this is misappropriating the text.

The ugly truth behind ChatGPT: AI is guzzling resources at planet-eating rates (www.theguardian.com)

Despite its name, the infrastructure used by the “cloud” accounts for more global greenhouse emissions than commercial flights. In 2018, for instance, the 5bn YouTube hits for the viral song Despacito used the same amount of energy it would take to heat 40,000 US homes annually....

Spedwell ,

The reason the article compares to commercial flights is your everyday reader knows planes' emissions are large. It's a reference point so people can weight the ecological tradeoff.

"I can emit this much by either (1) operating the global airline network, or (2) running cloud/LLMs." It's a good way to visualize the cost of cloud systems without just citing tons-of-CO2/yr.

Downplaying that by insisting we look at the transportation industry as a whole doesn't strike you as... a little silly? We know transport is expensive; It is moving tons of mass over hundreds of miles. The fact computer systems even get close is an indication of the sheer scale of energy being poured into them.

Spedwell , (edited )

concepts embedded in them

internal model

You used both phrases in this thread, but those are two very different things. It's a stretch to say this research supports the latter.

Yes, LLMs are still next-token generators. That is a descriptive statement about how they operate. They just have embedded knowledge that allows them to generate sometimes meaningful text.

Spedwell ,

It's not really stupid at all. See the matrix code example from this article: https://spectrum.ieee.org/ai-code-generation-ownership

You can't really know when the genAI is synthesizing from thousands of inputs or just outright reciting copyrighted code. Not kosher if it's the latter.

Spedwell ,

I get that there are better choices now, but let's not pretend like a straw you blow into is the technological stopping point for limb-free computer control (sorry if that's not actually the best option, it's just the one I'm familiar with). There are plenty of things to trash talk Neuralink about without pretending this technology (or it's future form) is meritless.

Spedwell ,

We should already be at that point. We have already seen LLMs' potential to inadvertently backdoor your code and to inadvertently help you violate copyright law (I guess we do need to wait to see what the courts rule, but I'll be rooting for the open-source authors).

If you use LLMs in your professional work, you're crazy. I would never be comfortably opening myself up to the legal and security liabilities of AI tools.

Spedwell ,

The issue on the copyright front is the same kind of professional standards and professional ethics that should stop you from just outright copying open-source code into your application. It may be very small portions of code, and you may never get caught, but you simply don't do that. If you wouldn't steal a function from a copyleft open-source project, you wouldn't use that function when copilot suggests it. Idk if copilot has added license tracing yet (been a while since I used it), but absent that feature you are entirely blind to the extent which it's output is infringing on licenses. That's huge legal liability to your employer, and an ethical coinflip.

Regarding understanding of code, you're right. You have to own what you submit into the codebase.

The drawback/risks of using LLMs or copilot are more to do with the fact it generates the likely code, which means it's statistically biased to generate whatever common and unnoticeable bugged logic exists in the average github repo it trained on. It will at some point give you code you read and say "yep, looks right to me" and then actually has a subtle buffer overflow issue, or actually fails in an edge case, because in a way that is just unnoticeable enough.

And you can make the argument that it's your responsibility to find that (it is). But I've seen some examples thrown around on twitter of just slightly bugged loops; I've seen examples of it replicated known vulnerabilities; and we have that package name fiasco in the that first article above.

If I ask myself would I definitely have caught that? the answer is only a maybe. If it replicates a vulnerability that existed in open-source code for years before it was noticed, do you really trust yourself to identify that the moment copilot suggests it to you?

I guess it all depends on stakes too. If you're generating buggy JavaScript who cares.

Spedwell ,

That's significantly worse privacy-wise, since Google gets a copy of everything.

A recovery email in this case was used to uncover the identity of the account-holder. Unless you're using proton mail anonymously (if you're replacing your personal gmail, then probably not) then you don't need to consider the recover email as a weakness.

'Vortex Cannon vs Drone' - Mark Rober shows off tech from a "defense technology company that specializes in advanced autonomous systems". That seems bad

I've enjoyed Mark Rober's videos for a while now. They are fun, touch on accessible topics, and have decent production value. But this recent video isn't sitting right with me...

Spedwell ,

I think it's more the dual-use nature of defense technology. It is very realistic to assume the tech that defends you here, is also going to be used in armed conflict (which historically for the US, involves in many civilian deaths). To present the technology without that critical examination, especially to a young audience like Rober's, is irresponsible. It can help form the view that this technology is inherently good, by leaving the adverse consequences under-examined and out of view to children watching this video.

Not that we need to suddenly start exposing kids to reporting on civilian collateral damage, wedding bombings, war crimes, etc... But if those are inherently part of this technology then leaving them out overlooks a crucial outcome of developing these tools. Maybe we just shouldn't advertise defense tech in kids media?

Spedwell ,

Wow, what a dishearteningly predictable attack.

I have studied computer architecture and hardware security at the graduate level—though I am far from an expert. That said, any student in the classroom could have laid out the theoretical weaknesses in a "data memory-dependent prefetcher".

My gut says (based on my own experience having a conversation like this) the engineers knew there was a "information leak" but management did not take it seriously. It's hard to convince someone without a cryptographic background why you need to {redesign/add a workaround/use a lower performance design} because of "leaks". If you can't demonstrate an attack they will assume the issue isn't exploitable.

Spedwell ,

I don't believe that explanation is more probable. If the NSA had the power to compell Apple to place a backdoor in their chip, it would probably be a proper backdoor. It wouldn't be a side channel in the cache that is exploitable only in specific conditions.

The exploit page mentions that the Intel DMP is robust because it is more selective. So this is likely just a simple design error of making the system a little too trigger-happy.

Spedwell ,

Having express self-checkoit is great. The Kroger near me went full-self-checkout. They have large kiosks that mimmic the traditional checkout belt kiosks, except the customer scans at the head of the belt and the items move into the bagging area.

If you have a full cart, you scan all the items, checkout, walk to the end of the belt, and bag all of your items. Takes twice as long as bagging while a cashier scans (for solo shoppers), and because of the automatic belt the next customer cannot start scanning until you finish bagging, or their items will join the pile of your items.

It effectively destroys all parallelism is the process (bagging while scanning, customers pre-loading their items with a divider while the prior customer is still being serviced), and with zero human operated checkouts running you get no choice

Spedwell ,

Yeah, I was complaining too. Or am I not understanding you?

  • All
  • Subscribed
  • Moderated
  • Favorites
  • •
  • random
  • incremental_games
  • meta
  • All magazines
    •