Google will catch up and top OpenAI in terms of AI product quality. GPT-5 (or whatever it will be called) will be initially thought impressive, but will overall perform only slightly better than GPT-4. On the hype curve, we will continue the downwards trend as realistic, practical applications will become commonplace. Nonetheless, VC money will still be poured into AI startups, hoping for another breakthrough which will not happen in a year.

oddly specific

In one year most people (including myself) will be disappointed they didn’t start an AI startup back in 2021-2024 due to the amount of low hanging fruit still available.

Those not inclined to make startups will be disappointed they didn’t spend the year learning about AI-related programming because those would be the top paying jobs, by far.

Those not into programming and not into making startups will be disappointed for having spent too much time on Manifold and other distracting websites instead of doing something meaningful with their lives.

Many more people try and fail to create some sort of "self managing agent" out of LLMs that can complete meaningful tasks, but this is actually several more years away at least.

Programming languages and tools that release in the upcoming year will start to experience more difficulty in gaining market share, as LLMs get better at speeding up work with existing tools, and LLM trainers become more hesitant to include recent data in their training set as the available online data gets polluted with ai generated text/code.

Negarestani's tellurian insurgency describes a perspective where oil is a demon freeing itself from the Earth. The moment humans first extracted oil from the Earth, we awakened a self-expanding entity, the only desire of which is to free more and more oil from the Earth. Humans like to think that it is their agency that causes them to pump oil, but in fact the more oil we pump, the more we become dependent on it, and the more we become the tools oil uses to extract more of itself rather than independent agents.

Over the next few years, we will see that AI will be the same thing. We are compelled further and further to develop better and more advanced AIs, and we will in the next year develop AIs that are not just chat bots but agents: AIs with function calling capabilities that will allow them to graduate from their current advisory role to becoming the decision-maker. Essentially, we will be further and further compelled to create the means of our own obselence. No doubt about it, it is AI that has the agency here.

AI Agents will begin to predominate over chatbots: the next wave of AI innovation will be in function calling, for example bioinformaticists using AI to run their genomic pipelines and software developers using AI to develop basic projects.

The time to finish a new project in software development from idea to deployment, which has already fallen from months to weeks, will be reduced to a few days.

It will become easier to create consumer-facing businesses where you put capital (gpus, energy, literal money) in and get recurring subscription revenue out. Eventually a venture capital formula will be developed for consumer apps, something similar to the institutional knowledge around SAAS - I don't think that knowledge will necessarily codify in a year, but some startup will have hit upon the flywheel by accident at least.

>>> What actions we personally should take today and over the next year

Step 1. Make more markets about AI
Step 2. Make even more markets about AI
Step 3. Profit

The fact that there are now some competitors to OpenAI for GPT-4 level models (i.e,Claude, Copilot, Gemini etc), is a major positive, also (Open Source) all working towards decentralization and spreading benefits more widely.

It also suggests that testers are now confident that GPT-4 models don't pose much of a threat (i.e., they're well below the level of AGI). It will be interesting to see just how much better the 'next gen' models are - i.e., Gemini 2.0, GPT-5. I'd expect these to come out either late this year, or early next year. They're probably still not AGI, but there's a small chance that they do turn out to be AGI with a bit more work - may be 5%.

The enormous compute costs and the unreliability of LLMs might mean it's time to look for a new kind of architecture that is much better from a safety perspective; something that's more reliable and allows for improved transparency and explainability.

It's likely that governments may start to get involved more to shape events; with unpredictable consequences. The motivations of the corporations is clear: innovation, profit , race for better AI. But the machinations of govts is a bit more mysterious . Folks need to bear in mind then, that corporations are not the only players, if they want to make sense of what is happening..

I don't doubt that progress in the labs continue at an exponential rate, and we are still on track for AGI within 8-10 yrs (2032-2034 ish), unless the powers that be decide to slow down.

If allowed by government it will show that the extreme left is lying and that Maga is not.