I'm bitter about the outcome too, but doesn't it seem like Paul could have _easily_ made a plan for the museum's continued existence or the donation of the computers into worthy hands? I can't see any scenario where the billionaire cofounder of Microsoft wouldn't have had those plans had he wanted them..so I can only conclude he _didn't_ care materially about the importance of the computers separate from himself.
I agree that it would have been trivially easy for him to give the museum a long-term future. I wouldn't jump to any conclusions about what he cared about.
Some might surprise you. I brought my Altair over to a friend's house and his clever 5th grade son was enthralled by it. I explained binary numbers and how to convert from binary to octal. Then I taught him some basic 8080 assembly instructions and he was entering his own programs on the front panel shortly later. I wrote out a basic program for adding two 8 bit integers and without my explanation, he figured out converting decimal <-> octal and chose some numbers you might expect from a 5th grader, like adding 6 and 7, and choosing two numbers that summed to 67 in decimal.
Achievement aside, the framing of this and the previous Claude achievement with the Jacobian conjecture don't sit right with me. They make it sound as though there is just some truly automated problem solving process underway at these major labs yielding novel math results on their own without human engagement. The OpenAI post [0] itself on this says "claiming human authorship for a proof generated entirely by an AI system would misrepresent both the system’s contribution and the nature of genuine human intellectual work." That's true only if there was _no_ human prompting involved to yield this result. In the case of the Claude result on the Jacobian conjecture, the employee who obtained the result _is_ a mathematician.
It seems to me that a mathematician is in a much better position than a lay person to construct prompts and engage with an LLM in a way that results in a new mathematical result. If he could work with Claude to obtain the counterexample but I cannot, that highlights the importance of the prompt engineering from the person, and their knowledge to do so. I'd be surprised that he was the first person to ever prod Claude to work on that particular problem.
Since they don't elaborate on the prompts or method that yielded the result, I'm unable to assume it was a fully AI driven approach. From that lens, I find the framing disingenuous.
Amusingly, I made the same thing late last year, though just on an ordinary computer, allowing you to draw in the browser. I used pageflip [0] with styling akin to Riddle's actual diary and a tiny local model crafted for roleplay via ollama. I remember writing "my name is Harry Potter, what is yours?" and getting back Snape, Malfoy, and even Harry Potter back across a number of iterations. After completing my experiment I learned I wasn't the first to think of this idea and found a few other similar AI Riddle diaries out there.
Incidentally, I have a Remarkable 2 and as of this weekend an m4 iPad air. Maybe I'll test this one out and see what the landscape for running models on iOS looks like.
I think attributing it to luck is a bit uncharitable. They described getting in on talent over credential. I also don't hold a degree. I've worked with a handful of Ivy League interns over the years and none of them blew me away. Certainly none were at the level I was at when I interviewed without a degree. I think I'm _fortunate_ to have gotten into the work without a degree, but it wasn't _luck_, it was ability.
it certainly has the tells... I could be charitable and say maybe since it's a bilingual blog it was generated as a type of translation from the original source, but that post itself seems only available in English
I feel it would be remiss of me not to mention the humble TempleOS support for mixing code and sprites. I can't say I've ever used the system myself, but its DolDoc format [0] offered some pretty nifty features for directly interspersing sprites into source code viewable together as one. You can see some screenshots on the hackmag page [1] under the section that reads: "If that’s not enough to impress you, open any game file and behold the unthinkable: the sprites are embedded right in the code!"
This was posted on hackaday[0] last week with a link to a youtube video. In the video, the author of the project goes into depth about some of the challenges they encountered and changes they made in good technical detail.
I think the complaints in this thread are not in the spirit of HN. Let's do better.
It's quite interesting, and describes how the developer used COBOL to implement a raycasting algorithm and generate a stream of PPM images to to pipe into FFplay. There's zero evidence of this being written in Claude; there are multiple clips of the developer working in VSCode, where the Claude plugin doesn't even appear to be installed.
The approach I've been trying to use at work is making heavy use of AI for the generation of experimental prototypes, but with little intention of keeping all that code verbatim. I can get something to the demo stage much more quickly than before, allowing me to show it to coworkers and users to get feedback on random possibilities. From there, I make a case to properly delegate time for "doing it the right way" as a slotted piece of work. That's resulted in a number of high-impact features being added to our app in the last few months, that in the past would have taken much longer or never came to fruition.
I'm using this specifically in the context of concepts/features that are hard to explain/sell without some working visual or prototype, but which aren't immediately evident needs or features requested by our users. Some of them go nowhere, but I think the net result is an increased ability for me to get my experiments from the "lab" to production.
reply