I see a pattern with AI companies. They always try to solve a really hard and not very useful problem. It's the same as with self driving car companies ten years ago: If you believe self driving tech is ripe for commercialization, the reasonable thing to do is something capital intensive and a special case where the technology most likely to succeed. For instance, heavy trucks automatically following others in formations for long drives. Saves gas, money, and potentially personnel.
There is a clear business case and buying large trucks is already a capex play. Then slowly work your way through more complex logistic problems from there. But no! The idea to sell was clearly the general problem including small cars that drive children to school through a suburban ice storm with lots of cyclists. Because that's clearly where the money is?
It's the same with AI. The consumer case is clearly there, people are easily impressed by it, and it is a given that consumers would pay to use it in products such as Illustrator, Logic Pro, modelling software etc. Maybe yet another try in radiology image processing, the death trap of startups for many decades now, but where there is obvious potential. But no! We want to design general purpose software -- in general purpose high level languages intended for human consumption! -- not even generating IR directly or running the model itself interactively.
If the technology really was good enough to do this type of work, why not find a specialized area with a few players limited by capex? Perhaps design a new competitive CPU? That's something we already have both specifications and tests for, and should be something a computer could do better than human. If an LLM could do a decent job there, it would easily be a billion dollar business. But no, let's write Python code and web apps!
AI allows for exquisite demos, demos that tantalize the audience into thinking of the infinite potential of the technology, that stunning vision expands and expands until the universe of potential overwhelms the dreamer into a state of terminal fantasy. So it is always a solution looking for a problem. There are cases where the two meet more realistically and a valuable impactful company develops it.
Agreed, the agents people are building are not solving the real issues.
The other thing people have been trying to do is build general agents e.g. Manus.
I just think this misses the key value add that agents can add at the moment.
A general agent would need to match the depth of every vertical agent, which is basically AGI. Until we reach AGI, verticalized agents for specific real
issues will be where the money/value is at.
That's exactly the approach to NLP which these super-successful LLMs are contradicting. They are generalists who can best with ease customized software developed over many years in all the subfields of NLP.
The railroad can’t have individual cars break off from the line to go to arbitrary warehouses, stores, and residences.
The railroad is an amazingly low cost way to move tonnage, if you’re going from a place where the railroad stops to another place where the railroad stops. There aren’t really companies that _could_ be using rail and aren’t.
But it just isn’t cost effective in many cases once you add in last-mile costs. If we built more rail (politically infeasible), you might see more usage but ultimately you still suffer from needing at least one locomotive per train.
I'm assuming this is a north america centric viewpoint - there are plenty of places in europe and asia where rail is far more common and has societal/political favor.
Solving the last mile by having stores that get shipments near a local train station that serves both cargo and passenger needs, and using kei trucks for small local deliveries is definitely a different set of tradeoffs.
It's a rhetorical question highlighting the contradiction of expecting unsophisticated systems that resort to follow-the-human tactics to be capable of self-driving inder more complex conditions in the last mile.
I have zero interest in debating the difficulty of freeway driving vs the last mile involves unmotorized participants and significantly more traffic control- that is settled in my mind.
Hah. Sort of. But the big difference is the railroad doesn't let anyone else use it. A regular road can support cars, trucks, truck convoys and maybe even bikes or pedestrians. A railroad can support trains.
Do they, pragmatically speaking? High-volume cargo traffic quickly wears down the asphalt and causes regular jams, a bike lane unseparated from cars is a safety hazard enough large enough to push many potential riders off the road, and most morning commutes would be better served by well-developed public transit.
One EMD SD70ACe locomotive moves over 10,000 tonnes of cargo using 1,300 L of diesel per 1,000 km. The equivalent 286 trucks would consume 107,250 L, while needing 55.8 km of a single-lane highway, compared to the 2.16 km freight train.
Similarly, the average US car has 1.5 passengers per ~30 m² of space, so 20 m² per person. An average bike is about 2 m² per person. A typical trolley car holds ~160 passengers per 200 square meters, so 1.25 m² per person. A tram reliably moves at 60–80 km/h on interurban routers, or 30 km/h in urban centers with frequent stops, a considerable improvement over San Francisco's 16 km/h by car for last mile.
the problem with rail is that it's hard to scale up (and down)
getting new tracks built takes waaay too long (because of NIMBY and simply because the road is usually already there)
there's no long-term thinking from politics, and no market forces converging to somehow over the years lead to some compounding (so the inefficiencies don't really translate to some big problem -- well, climate change and slower GDP growth)
There is a clear business case and buying large trucks is already a capex play. Then slowly work your way through more complex logistic problems from there. But no! The idea to sell was clearly the general problem including small cars that drive children to school through a suburban ice storm with lots of cyclists. Because that's clearly where the money is?
It's the same with AI. The consumer case is clearly there, people are easily impressed by it, and it is a given that consumers would pay to use it in products such as Illustrator, Logic Pro, modelling software etc. Maybe yet another try in radiology image processing, the death trap of startups for many decades now, but where there is obvious potential. But no! We want to design general purpose software -- in general purpose high level languages intended for human consumption! -- not even generating IR directly or running the model itself interactively.
If the technology really was good enough to do this type of work, why not find a specialized area with a few players limited by capex? Perhaps design a new competitive CPU? That's something we already have both specifications and tests for, and should be something a computer could do better than human. If an LLM could do a decent job there, it would easily be a billion dollar business. But no, let's write Python code and web apps!