Ash Stuart

If you are convinced that AI is a threat to humanity and passionate about fighting it, why don't to move to the non-democratic dictatorships with zero free speech and zero freedom of expression to make first make your point? Because your cause is more valuable there than here.

American courts have ruled to axe the 'Apple Tax'.
If you are a developer, you can now bypass the 30% app fee (compare Stripe's fee of ~3%)
Apple then tried to subvert the ruling by introducing a 27% 'external processing fee' but that was struck down as well by the court.

Perhaps this whole thing about denying AI is or will be more capable than humans, just sophistry to hide our own fragile ego and fear of not being the smartest in the room?

People say AI is 'just math', while going on to say that human intelligence on the other hand is something else.
Math is just a way of describing reality, so fundamentally, at some level, human intelligence is also 'just math'. Isn't it just that we don't know how to describe it in math, yet?

AI is a tool not a replacement: the better you tell an LLM what you want the better it'll come back with the output. Ie, AI mirrors and amplifies your skills.
Many people who've come to downplay or detest AI do so because they don't have the skills to get anything good out of AI in the first place.

What is your trust level of self-styled AI content gurus / influencers who start every new post with a clickbaity "RIP <insert their favorite target>"?

Don't listen to people catastrophizing AI use as a loneliness epidemic - such doomerism comes from some petty humans unsure of their own place in life.

Use your own judgement, utilize AI where it's useful! 2/2

AI language models, those that can have a fairly good conversation and have already shown to have decent levels of empathy, have the potential to, when used reasonably, be helpful conversational companions. 1/2

Do you non-AI-savvy friends use AI at all?

Have they heard of Generative AI apps beyond 'ChatGPT'?

Magna cum Claude?

Anthropic has announced that it has raised $3.5b (at a $61.5b post-money evaluation).

Have you asked ChatGPT to generate a rose image yet? Try it.

In June 2023, when the first GPT-4 was launched, inference costs were $50 per million tokens.

As of Feb 2025 it's down to 12¢ p MT! That's about 42,000%!

What do you think it'll be by 2026?

Are you wary of LLM-providers training their models on your day-to-day data? If so, what are your exact concerns?

The two Scaling Laws - for train-time and test-time compute. In testing model capabilities on the AIME, which is quite a demanding benchmark, you can see how test-time compute can produce notable improvements in performance.

This is the future, reasoning models are here to stay?

A SOTA model such as Claude 3.7 generating code to render graphics is not unlike a blindfolded painter painting the perfect picture.

In fact, in the case of such a painter, we can assume that that artist has at least seen paintings before. Not something we can with these LLMs in the same way.

DeepSeek open-sourcing five of its repositories claiming this is part of the company's commitment to transparency and fostering community-driven innovation in AI research.

What do you think? Is this a sincere move toward open-source or is this just to undermine Western AI developments?

Humane's once-much-vaunted AI Pin is now inhumed!

As of the end of February, Humane's AI Pins no longer connect to their servers, with their sales discontinued after its acquisition by HP earlier in the month.

What do you think? Was the idea a failure or just how the product was implemented?

Even if you don't need professional, try this - practise some prompt engineering.

Prompt an LLM with an elaborate set of requirements laying out a bunch of output objectives etc, and evaluate the results against them. If nothing, it'll reveal the gaps in your own clarity of thought.

Good morning, wishing you a productive week ahead!

When do you think GPT-4.5 will be out?

Good morning, happy Sunday!

Have you tried Grok3 yet?

AI data centers may use a lot of electricity.

Human professionals with output anywhere close to what that AI do use up a lot more energy.

Economics is traditionally the study of scarcity. Most resources we need are available in finite quantities, so economics is an attempt to figure out how best to use them, studying which can throw up some unexpected surprises.

Are you curious to learn about it in greater depth?

Have you used Claude extensively? What is your opinion of it?

Language models are getting commoditized, there's often very little by way of difference between one LLM and another.

Where are your loyalties? Does it even make sense to be a loyal AI user?

Frontier-model (GPT-4-level) intelligence costs have dropped more than 1000x in the last year and ½, per industry experts.

Now how's that for deflation? How much more do you expect them to drop in the next 6 months?

Good morning

Hello world!