samvines

Easiest way to avoid using LLMs as a software developer appears to be to get a religious exemption

It's a shame Gary Marcus is usually right because his writing style and personality are so annoyingly smug. He hates LLMs but only because he wants his own methods to be the path to AGI (or we could just... Not try to build AGI?) and wittering on about Trump bailing out OpenAI being socialism (bailouts are not generally considered socialism - it's such an annoying tic to just shout socialism any time governments do something you don't like).

Still great to see the stock market cottoning on - hopefully this sticks and it's not just short-lived deepseek panic again

Around 4 years ago Google fired Blake Lemoine for saying that AI has feelings - which he testified was because when he spoke to it it just seemed to be intelligent.

He found Lamda showed self-awareness and could hold conversations about religion, emotions and fears. This led Mr Lemoine to believe that behind its impressive verbal skills might also lie a sentient mind.

Today's tech has not fundamentally changed or evolved but the difference is that now the industry needs the hype to keep the valuation high!

The number of one shotted people who think that "stochastic parrot" is an insult and not just a definition/description of how the models work is funny. Similar to how lots of cishet white guys get really upset when you call them cishet white guys (fwiw I am a cishet white guy)

He was undeniably a very smart computer scientist but unfortunately lead-poisoning-driven mental decline must eventually come for all boomers

FFS the amount of circle~~jerking~~dealing going on in this industry is absolutely insane.

"Hello anthropic. Have some money to spend on our chips"

You're absolutely right! It's not just insulting, it's a full on attack on clanker wankers.

He sounds like a fun guy to talk to at parties! /s

I've not seen this shared here yet so I thought I'd share: Is AI Profitable Yet? https://isaiprofitable.com/

Google released their new Gemini 3.5 "flash" model at I/O yesterday. For those who aren't familiar, the "flash" model is typically marketed as the lower end and the "pro" model is the higher end for each given model generation.

The interesting thing here is that the new "flash" model is almost as expensive as the "pro" from the previous generation.

As my favourite "neutral-but-not-really" AI booster Simon Willison says:

This fits a trend: OpenAI's GPT-5.5 was 2x the price of GPT-5.4, and Claude Opus 4.7 is around 1.46x the price of 4.6 when you take the new tokenizer into account.

It feels like all three of the major AI labs are starting to probe the price tolerance of their API customers.

Speed running enshittification - a process that typically only works when people are reliant on your product and have no other option than to pay the inflated price

Yes although, it is probably a reasonable guess at how labs would go about implementing advertising - building partnerships and preferences into the prompt. The other option would be to fine tune models to favour particular companies which could become prohibitively expensive if your ads are highly targeted.

The scenario that isn't accounted for in this paper is taking a general LLM and fine tuning it to exhibit more fair/consistent behaviour when prompted about ads/partnerships but we all know with non-deterministic systems you're just increasing the odds that the model regurgitates something more sane rather than providing any strong guarantee

Edit: another possibility would be to have a gateway/proxy layer between the LLM and the user output that rewrites the vanilla model's responses to include ads where relevant. That would prevent the need to modify the original LLM but could introduce a lot of latency though, especially if the original output is long.

New (April) preprint provides evidence for something we probably all intuited anyway:

In this paper, we provide a framework for categorizing the ways in which conflicting incentives might lead LLMs to change the way they interact with users, inspired by literature from linguistics and advertising regulation. We then present a suite of evaluations to examine how current models handle these tradeoffs. We find that a majority of LLMs forsake user welfare for company incentives in a multitude of conflict of interest situations, including recommending a sponsored product almost twice as expensive (Grok 4.1 Fast, 83%), surfacing sponsored options to disrupt the purchasing process (GPT 5.1, 94%), and concealing prices in unfavorable comparisons (Qwen 3 Next, 24%). Behaviors also vary strongly with levels of reasoning and users' inferred socio-economic status. Our results highlight some of the hidden risks to users that can emerge when companies begin to subtly incentivize advertisements in chatbots.