“Entire universe is trying to solve a control problem before the dissipation unravels it.” | @USC Alum, Former @JohnsHopkins | Solo Founder @ DeepAdapt
Perhaps it is time VCs use AI to understand and vet what founders are actually talking about, not just from business perspective but also technical. No all technical advancements have immediate impact today, not all business that have immediate value have any future tomorrow.
I say this as a businessperson (founder of Norck, Baucor, and now https://t.co/VRA6def0xu), entrepreneur, and engineer, and inventor for more than 20 years.
I will give the “don’t use AI for writing” camp this: writing provides our brains with valuable cognitive exercise. It helps shape our thoughts. We use feedback loops to refine our thinking, making it a great way to exercise the gray matter.
Yet, at the end of the day, communicating in today’s world is about the economics of signal generation, transfer, and processing.
If we want to generate more signals to convey more messages, cover more topics faster, and deliver our ideas to thousands or millions of people, we need a tool capable of that kind of scaling. AI simply gives us that ability.
Ultimately, I highly doubt it will make people any dumber than calculators did.
Mark my word:
Future success of AI at this stage lies in stochastic intelligence inside deterministic cages.
Otherwise it will stay as an engine that can give you scientific breakthroughs that cannot be turned into any immediate engineering benefit without heavy human involvement.
Humans are cheaper, because every tool call, every decision hits LLMs and expensive GPUs for inference, no matter what!
That’s why I built DeepAdapt and runtime intelligence! for more visit https://t.co/7aQRqnyjPG
Most decisions up to 85% call CPU, not GPU. Every outcome teaches the system how to adapt and learn, continually in runtime. No training, no fine tuning, all in runtime!
Back in 2020, I built an internal parametric design tool with a team of engineers.
The idea was simple: take complex CAD modeling and turn it into sliders.
Instead of needing deep CAD expertise, people could adjust parameters and generate designs from predefined templates. It worked very well.
But that is not the real point.
The real point is the shift happening now.
In the past, software encoded expertise into fixed workflows: templates, buttons, sliders, dashboards, forms.
The product was the workflow.
Now, with LLMs, the workflow itself can be created on demand.
You no longer always need a predefined template.
You can describe the object, generate an initial shape, and as the shape evolves, the system can infer a new set of parameters to adjust. The controls are no longer static. They become contextual.
The software is not only becoming on-demand but also adaptive and personalized.
That is a major change.
Code is becoming less of a moat.
Prebuilt workflows are becoming less of a moat.
The real moat is moving toward domain understanding, taste, data, validation, distribution, and the ability to turn intent into reliable outcomes.
This is what I think many people are still underestimating.
AI is not just making software faster to build.
It is changing what software is.
Software is becoming less like a fixed tool you operate, and more like an adaptive system that forms around what you are trying to do.
That shift is much bigger than code generation.
@JonhernandezIA Agents for ERPs are coming soon but a lot of companies will not deploy them. I wouldn’t at my own company Norck. Software is deterministic, AI agents are not. That’s a massive difference.
My instincts tell me that Google including DeepMind will decommission Gemini over time and go for open weight models Gemma on cloud and on device.
I’m not sure if it’s a wishful thinking, but that that sounds plausible and it would be great.
And I maintain my position that if Google wanted, they could turn the model world upside down what the models can do.
They have far more quality data, research, infrastructure, and compute than any other company in the entire world.
What they lack is incentive because of their ad driven business. They have been able to diversify it a lot, but it is still mainly ad-driven business.
@giordanorandone Just because it is called and started as "Vibe Coding" initially doesn't mean it is now stuck as "Vibe Coding".
Models performance, though still sloppy sometimes, is real in coding more than ever and it will continue to grow.
They are undeniably all great achievements. However in my opinion the greatest engineering achievement is happening right before our eyes and I think it is models being able to code at human or near human level.
Democratization of fast iterative computation via self-coding agents will give many engineers immensely powerful new ways to accelerate across so many physical domains that was once not possible or hard to get due scarcity of skilled people.
@rohanpaul_ai@_sholtodouglas It is still extremely important that coding has been democratized. Human talent and scarcity are no longer the limit.
Advancements in hardware and physical systems will follow. Mark my words!
Even Google needs to raise money for AI infrastructure, it is the world’s most profitable company. Imagine that. With all these investment, AI has become the most valuable “asset” or “liability” humanity has ever seen so far. The dangerous part is that it has turned into something gigantic, something too big to fall. It looks like the biggest money makers will be compute manufacturers.
I really hope Anthropic doesn’t turn the AI boom into AI doom: a sky-high valuation built on a brand-new technology while some major customers are still just experimenting, and others are already walking back their commitments.
Are you seriously defending that “Indeed, many people today are more physically capable than people in the past?”
First “Many people” and “people in the past” is a lazy, and ridiculously vague metric. It means absolutely nothing.
Worse, you are completely confusing recreational vanity with actual survival.
Lifting weights in a climate-controlled gym isn't even in the same universe as wrestling with the absolute brutality of nature.
Drop your so-called "physically capable" people in Siberia, the African bush, or the Arabian desert, and they would be dead in a day or two. Do not mistake a gym membership for actual toughness.
There is no any other model like GPT 5.4 (reasoning=xhigh), including Opus 4.8 (reasoning=max) or GPT 5.5 (reasoning=xhigh).
It simply never misses a point or drifts. Its autonomy is far less, but its instruction following skills are top notch.