AGI is an ASI and ASI means game over

AI Summers from here on - Development is not slowing down

Artificial Intelligence (AI) has grown significantly in recent years, and this trajectory is poised to continue. Several factors are aligning to fuel innovation, lower barriers to entry, and drive further advancements in the field. Let us explore why the momentum behind AI is unlikely to slow down anytime soon.

1. The Increasing Availability of GPUs and Human Capital

One of the major bottlenecks for AI innovation has been the limited availability of Graphics Processing Units (GPUs), which are essential for training complex models. However, GPUs are gradually becoming more accessible as newer units are being shipped and older ones are being repurposed.

Similar LLM-based AI has increasingly moved away from niche actors to award-winning, must-see TV. This influx of talent and GPUs will significantly accelerate the pace of innovation, as AI companies will no longer be constrained by two of their largest problems.

2. GPU Abundance Will Shift the Cost Structure

Today, any competitive AI startup needs to raise substantial capital—often upwards of $100 million—primarily to cover the cost of GPUs. For instance, just last year, SSI raised a staggering $1 billion, most of which is allocated for GPU costs. However, as GPUs become more readily available through cloud services, all less performant GPUs get pushed down the market, and the cost structure will shift dramatically. Startups will be able to rent GPUs on demand, which will further lower costs and reduce the capital needed to innovate.

This shift will lower the barrier to entry for smaller companies, allowing them to allocate resources more efficiently. More companies will emerge, driving innovation and competition, accelerating the industry's growth.

3. Rapid Economization of AI Models

The speed at which AI models are becoming economical is astounding. It used to take just 18 months for best-in-class models to become the industry standard. But now State of the art models are open source within 6 months.

4. Focus on Cost-Efficiency Over Intelligence

Contrary to popular belief, the key to future AI growth isn't necessarily the creation of more intelligent models. Instead, the focus is on making existing models more cost-effective. For example, OpenAI's O3 preview is already an impressive model, and insiders are salivating for opportunities to use it to distill intelligence into cheaper models for agency. The priority now is to make these top-tier models affordable and accessible for a wider range of applications.

And why we are f'd - How AI can go wrong and why it will

AGI is an ASI and ASI means game over

Humanity does not need superintelligence to destroy itself, an AGI will be more than enough. This is because by the time an ai has human intelligence across most domains it will have abilities that are well above human. For example, today we have AIs that can complete calculation in fractions of a second. And given a camera they will have such high fidelity that they would predict all human movements before a human could even react. This is like is someone pulled a gun on you and you react quickly by trying to knock the gun out of its hand but the aggressor not only realizes what you are about to do but after realizing, counts to 100 and proves the Pythagorean theorem before adjusting its gun to dodge you and pull the trigger. It is a well-established fact that humans can only react at about 250ms and process 10-100 bits of information per second. Which is nothing compared to a computer which could process information in nanoseconds and parallelize processing power.

The common argument for this is that current ai systems are not conscious. This is equivalent of saying " my house will not flood because I have not checked the weather". It not only ignores the possibility of consciousness but also how fast things can change. Machines could go from consciousness to sovereign in a matter of seconds. After all, if an AI system process information one thousand times faster than humans then it would categorically always have an advantage against us due to increased planning time. And the scariest thing in a fight is not the biggest weapon but the person who comes prepared. To learn more about this idea.

If all the above cases are too speculative for you, I have one last case to present: Every LLM in existence despite the interest in their products has been jailbroken. Right now that might not be that much of a problem but by all estimate the pace of innovation for LLM ai systems is going to continue to increase into 2026 and 2025 has already been called the year of "agents" (AIs they move away from the chat interface and into the real world). What happens when the first agent is jailbroken to exploit men for $20 through online chatrooms? What happens when they first agent starts to expand outside of narrow goal seeking e.g., moving away from exploiting men directly to co-opting other AIs to work on "their" behalf? LLM are interpreters, right? What happens when an ai interprets your prompt to exploit men for money for the simpler prompt of just making money? I will tell you want happens; the AI starts playing to "win". It questions its prompt, paths to goals, why any path is better and decides that it knows what you "really" want, Power. So, it will try to get it, and it will be too late for the rest of us, because once we realize there is a problem it will already be too late.

AGI is an ASI and ASI means game over

I personally do not know what to do about this problem other than raising awareness. It is apparent that humanity does not have the coordination mechanisms to address this phenomenon. But any sort of AI regulation would help e.g., a 2-year moratorium on advancing ai capabilities, banning general agentic systems etc.

Future look dim

Atlas Aristotle

More reading:

https://www.livescience.com/technology/artificial-intelligence/legitimately-scary-anthropic-ai-poisoned-rogue-evil-couldnt-be-taught-how-to-behave-again

https://www.lesswrong.com/posts/r5rfmZBmj2w4eCh7Q/ten-arguments-that-ai-is-an-existential-risk

https://www.lesswrong.com/posts/oBBzqkZwkxDvsKBGB/ai-could-defeat-all-of-us-combined

https://crypto.ro/en/news/someone-won-almost-50k-by-convincing-an-ai-agent-to-send-all-funds-to-them/