Post

Anil Dash
@anildash.com

The thing about this, as someone who’s very versed in LLMs and related technologies, is that the mainstream, mass-marketed versions of “AI” are *worthy* of contempt! Even if you like this tech, why would you advocate for such shitty versions of it? The Internet didn’t *start out* being shitty!

Marisa Kabas@marisakabas.bsky.social

one interesting thing so far about loudly hating AI is the way AI lovers are like "wow so you're just totally closed off to hearing another perspective??" as if their perspective hasn't been relentlessly pushed down our throats with no space to push back (particularly in professional settings)

130 reposts668 likes

Anil Dash
@anildash.com

Like, if you believe that LLMs are a powerful technology that could be useful, that’s a big reason to be ferociously critical of the Big AI companies! Just like loving the internet is a good reason to hate Meta/Facebook. Why the hell would you root for big evil companies making the worst version?

Read Hardt and Negri
@jason-edelman.org

What would you propose? Inference as a public utility? A global AI federation? We're cranking away on the open weights models and local inference but it's hard to go toe to toe with like, the entire capitalist superstructure

Anil Dash
@anildash.com

I think a multiplicity of approaches — part of the problem itself is trying to have one answer for everything. So, all of those, but also academic offerings, lots of public options being created in the Global South, purpose-specific open models in particular verticals. Community-based answers.

T
@hillhand.com

The Google AI overview being unable to give a definition for "disregard" is the perfect example! It's *possible* to use an LLM to do work, but it does *not* look like just hooking up an LLM directly to your user.

Anil Dash
@anildash.com

Ironically, I think in a lot of scenarios, the best way to to use LLMs is for them to prompt capable humans, rather than the other way around.

Craig Thomler
@craigthomler.bsky.social

I agree that LLMs benefit experienced and educated people the most - because they can guide the AI and pickup the falsehoods (mostly). What is particularly bad is that education hasn’t caught up - so the gap between experienced and inexperienced is growing & pathways to experience are narrowing.

Craig Thomler
@craigthomler.bsky.social

Recognising they exist is not the same as advocating for them. If you have a method of practically removing them from the world, please share it. Otherwise any pragmatic strategy has to factor them in. Don’t confuse recognising reality with finding the present situation desirable.

Ryn
@reprapryn.bsky.social

Modern small models are quite nice. Like a skilled person with a pretty bad ass desktop running an open model and the right tooling is going to outperform a medicore dev spending lots on claude But the corpos want the latter, because its rent seeking. With a side of ending that pesky middle class.

Anil Dash
@anildash.com

The Internet was far more than that pre-web! Usenet thrived for decades, mailing lists were wildly valuable, and FTP enabled massive innovations. Even Gopher unlocked huge amounts of information sharing. I still used all of those for many years after the web showed up; we’re using text right now.

Ghostly Bree
@mostlybree.kitrocha.com

the early internet was *incredibly good* at being what they designed the early internet to be. FTP transferred files from one place to another. Mailing lists sent mail to a list of people. I do not feel the same way about the current crop of AI.

Ghostly Bree
@mostlybree.kitrocha.com

If you'd booted up FTP and sometimes it transferred your files to where you told it to, sometimes it transferred them somewhere else, and sometimes it deleted them? There would be some understandable backlash!!

Craig Thomler
@craigthomler.bsky.social

I used it too. It was shitty compared to what we can do today. No system starts out perfect. Take off those rose coloured glasses before you start telling us how great society was in the 1840s.

Anil Dash
@anildash.com

It is fucking wild to see a white man condescend to tell a person of color that things are better now than before the Civil War. Things were better before you jumped in my mentions and tried to lecture me on topics that I know far better than you.

Jon
@jontafkasi.bsky.social

Things that are good don't need massive tech companies to relentlessly market them, force them upon us when we didn't ask for them, include them in what are otherwise half decent products (and now aren't), and provide no or hidden options to turn them off.

Ryan
@coverthistory.cryptoanarchy.network

Not only does it guess that is all it can do. Its incredibly good at it but it maps how word relate to each other in a modeled space and then guesses the next most likely token to generate.

Stephen James
@stephenjamessv.bsky.social

I don’t know the technical terminology, but when you call it out for giving you bad info, Google search AI is also good at giving excuses that sound maybe feasible but probably are just made up excuses like “failing to read live stock sheets correctly.” How does that happen?!?

Rusty Foster
@rusty.todayintabs.com

When you talk to an LLM, you and the computer are writing a little improv scene together. It’s just playing in the narrative space with you. Whatever you give it, it responds in the way most likely to keep you engaged. It’s only ever informative by accident.

Ryan
@coverthistory.cryptoanarchy.network

A way to cut through that a bit is to reject its apologies. Say something like "You have no agency or being, and and are not experiencing regret.Your apology is a simulation of human emotion for the purposes of manipulation and inappropriate"