Better plan is to shut down PhD and make students take a oral thesis at bachelor and master level and help them become a productive economic participant as soon as possible.
As a startup with deep pockets, you think you can afford to be generous, without realising it is made possible due to borrowed time and money. When the bills come due, generosity ends and enshittification onsets.
People are more sympathetic to pg retaining a sense of purpose (than Terry Tao) because they rightfully imagine that's it's easy to obtain the virtues that he has (the generosity, the optimism, the agency,the lack of worry, the curiosity, the faith in all sorts of humanity)
Newsflash. Most graduate research is worth nothing. If this helps graduate students to re-evaluate and start doing things that can generate income while being intellectually stimulating, it's a net win for the person and the economy
I mean it is. Most of the time in graduate school, you're getting trained and learning to do reasearch. People rarely produce great work during graduate school. Only towards the end of their degree or start of new position (post doc, independent researcher, or assistant professor), you start seeing good work. And for a professor, most of their research is done by graduate students.
Academia has it's own sets of big problems. Before this AI boom also, most people graduating never stayed in academia. The way graduate schools are structured, I would be happy if few people joined. That would also mean we lose good researchers in a long run.
I like the hacking spirit suggestion. But my disagreement is the specificity of it. The suggestion here seems more inline with not tinkering and hacking but first principles learning typical of science and math curriculum. What made hacking radios and JS cool was the short feedback loops to see the progress you made. Unfortunately for LLMs starting from scratch doesnt give a good feedback loop on improvement, although starting from pre-trained model to quantising, fine-tuning might be more aligned with the hacking culture
Good to know other people who get migrane reading LLMs dense prose. I started reading books again recently, since everything online is polluted by LLM prose. What i realise, is that a human author, especially a teacher understands the learning pathways of new learners, they motivate the learning, and start from simplest concepts (a spherical cow), and then building all the complexities. This helps us to emphasize on most important concepts, while throwing away unnecessary complexities. While reading LLM prose is like reading a research article, that is written to an expert in the area, that talks about bleeding edge, with full of jargons, caveats, that just is not conducive to the learning process for a new learner.
Also they tend to assemble complex jargon in obtuse or meaningless ways, which makes reading and parsing and understanding much more difficult. Tends to reveal that LLMs fundamentally do not have "understanding", just likely word generation
To me, it's just Claude. The other models have their quirks but nothing is quite like Claude.
But even with Claude, it's it's really the prose getting in the way you can install the caveman plugin or tell it to use that "standard technical English" thing.
I wish I had something more methodical I could show. It's all subjective, but GLM-5.2 feels more human to me. Even GPT-5.6 Sol tends to be easier on the eyes for me (though the stereotype of it overengineering and no common sense are still true).
I tried using a new agent service recently and could tell immediately that it's powered by Claude due to the way it writes.
A slight tangent here is a technical deep dive on Voyager 1 fixing a memory corruption error in 2023 by hand 15 billion miles away. Blows my mind of the reverse engineering done to achieve this
https://www.youtube.com/watch?v=YcUycQoz0zg
Voyager program and the team behind it will always serve as an example for me. Over-the-space software updates, 50 years support life. Debugging and patching devices with no room for a mistake. How many of 2026 devices connected to internet will have that luxury in 2076?..
100 percent can relate on the relentless curiosity of a child. I was dumbfounded yesterday by this question -> if cars go at high speed, and breaks stop it suddenly, how come the friction doesn't generate enough heat to start a spark?
They generate enough energy to, I mean have you seen racing brake discs?
It's just that the energy is spread out over a larger area (brake discs/pads as opposed to a tiny spark) that is also constantly being cooled by air movement, so the energy in any one point is not enough to ignite any of the materials involved. It's why they used to put asbestos in the brake pads, but they now use ceramics and other heat-resistant materials.
I'd approach it by looking at the definition of a "spark", and the things we can change to avoid tripping that definition.
For a visual spark, it has to be something thrown off which is the right size, material, and a temperature to glow visibly to the human eye under normal light.
For a spark that starts a fire, it has to be something thrown off which has a certain temperature by the time it lands or travels to some other surface.
In both cases, hot very fine particles will cool quickly just by hitting the ambient air. If the underlying surface (e.g. brake disk) gets glowy-hot, we wouldn't call that a spark--it's too big, and fixed.
In my experience sparks are generally generated when quite small metal surfaces get in contact at high speeds, meaning they can generate a lot of friction in a very small point, and thus not have enough braking effect despite the local friction in one small point (think rail wheels squeeking in curves).
Brakes OTOH are designed to spread out the friction on a somewhat larger area, which allows more material to take up the friction and the heat, and spread it around.
I know brakes can get extremely hot too though, and after a really hard braking you might have to worry about deformed brakes sometimes.
Race cars tend to have really large breaks (and brake pads) for this reason.
Car brakes can get hot enough to start a fire (wood, paper, maybe even gasoline - see race cars) but a spark is a hot piece of material that is large enough so it can keep its heat for a while.
Brakes generate very fine dust that cools off almost immediately.
The concentration of energy is vital. Sort of like how if you fire a gun it will only push the person back a few centimeters but that energy condensed into a very small spot and fly through a car.
I would answer that heat is a form of energy. A horse pulling a carriage a 1000 yards requires more energy than pulling 30 yards.
When a car slams on the brakes the distance is too short to generate enough heat to deform. I would also suspect the treads would partially melt and deform before it catches on fire.
This is supercool. But I have a tangential question. I see projects like this built on top of OSM for different use cases. Does this data gets contributed back to OSM. A lot of them seems hobby projects, so it's hard to maintain long term, but the data is quite useful and could be used beyond the lifetime of these projects if they can be contributed back to OSM somehow.
OSM probably has the best data on golf courses because for many courses, people have mapped each 'golf=hole' as a way, and mapped other golf course features such as 'golf=bunker', 'golf=green', etc.
OSM knows of almost 40k golf courses around the world.[2] This compares to 38k golf courses known to the R&A Golf Around the World Fourth Edition (2021) report and 37.6k courses listed in the NGF course database as of July 2024. It's likely OSM knows of 99.99..% of all active golf courses and also still retains tags for closed/unused golf courses too. So if OSM were to be inaccurate, it'd likely be inaccurate for whether a golf course is open or closed, rather than OSM not knowing about a golf course existing.
I wonder whether an OSM tile server / custom style sheet for golfers would be easiest and most useful/comprehensive way to keep track of golf courses?
Looks like my approach is as old school as it gets. For complex piece of work, the only way to be in flow for me is when I'm driving the engagement. I start with a high level requirement and a high level design plan to achieve it. I also provide constraints that needs to be satisfied (efficiency, performance, cost, scale etc) and write it all into a markdown document and ask LLMs to review it, find blindspots and refine it until I can get a detailed design for that phase.
Then I pass that to another LLM provider to review and check for any bloat that can be cut or blindspots that need to be addressed.
Finally I get a test plan to help me test different components directly and in debug mode. I then ask LLMs to implement in stages where I can test them in small components as possible.
I think I end up spending more time (easily 2x) than hand rolling. But the upside is the design is more thought out compared to hand rolled code. It has fewer accidental complexities and I have a clear mental model of the entire design that can also be shared to others through the document
reply