It gets complicated, because protons and anti-protons are composite particles.
To simplify somewhat, one quark from each will annihilate, leaving four quarks and a lot of energy, which re-arrange into mesons (two quarks each), which then decay into photons, electrons, positrons and neutrinos.
Annihilation. At low collision energies the antiproton and proton go through an interaction which ultimately results in photons (light energy). Matter is converted into energy as e = mc^2. Though the interaction passes through a number of exotic intermediate stages first.
No, I understand. But if you have a pair of arms and hands that can't do that long tail, why add the legs? Especially when you know others have already gotten legs to work, but no one has made a robot that can just do any task in arms reach.
Most tasks would me more efficient if not using a drop in replacement for humans. Rethinking the existing paradigms and redesigning the processes to be more efficient with non-humans would be more valuable.
It may be true that it might be more efficient, but it feels similar to the farming paradigm. You can actually get more calories per acre if you have ecosystems with varying heights. However, we don’t do this because it is much more scalable to drive massive machines over flat, unimpeded land.
It might be the case that my home would be more robot-friendly if I lived in a flat warehouse that was designed so that robots could easily handle tasks. But I don’t want to live in a warehouse, and I don’t want my house to be optimized for robots. And this isn’t even to mention the long tail of existing houses that do not adhere to a robot-optimized standard. The cost of replacing or retrofitting all of them would be astronomical.
"However, we don’t do this because it is much more scalable to drive massive machines over flat, unimpeded land."
It's just that it's cheaper since we're talking about land and other physical constraints. With humanoid robots we're talking about labor. Many processes done by humans are designed to be possible by humans. As soon as you remove the ability constraints of humans, you can design more efficient processes. This is evident is all jobs pre- vs post-mechanization.
For example, do you need lights in a warehouse without people? Do you need maximum shelf height or forklift speeds for safety standards? Do you even need forklifts if the bots are stronger enough; if not, aisles can be narrowed and density increased. Etc.
Even for in-home assistance, you might be better off with multiple smaller bots tailored to different tasks. A bipedal robot designed for cleaning gutters or shoveling snow sounds like a poor design decision.
It is not worth a lot. At the very least, not a single robotics has proven this by demonstration despite billions in funding over the last many years. In fact this is also true for a lot of non-humanoid "ai" powered robots.
The models that are really good at 3D modeling tend to also be really good at pelicans. I think the skills are strongly related - it's about having good 2D (and now 3D) spatial awareness.
The test is always going to be hard though, because it's an absurd image to ask for.
Models seem to have very interesting blind spots. While https://assbench.com/ may seem like a joke, the results are very illuminating. One might expect far better results given the detailed prompt, allowance for render checks and iteration, the number of human figures the models are trained on, and clarity of tool use (threejs).
However, result quality really drops off once you get out of the top 5 or so.
(NSFW note: all buttocks are covered by fairly conservative bikini bottoms and show no genitals, bulges, etc.)
Yeah, that's a good point. My basic pelican test doesn't allow the model to render their work and iterate on it. The 3D models all come from agent systems that are rendering with Blender or screenshotting with Playwright along the way.
Have you looked at assbench.com? Even with threeJS rendering, screenshotting with Playwright, and iteration (check the prompt) some of the models are shockingly bad. I wonder whether the rendering quality is affected by the requirement for skin physics on click; if so, it raises a valid question about models creating pretty pictures but break when interactivity is required.
(As I said in a separate comment above, results are all bikini-clad and not revealing. However, whether it is NSFW is up to you/your work policies.)
I’d really like to see what each model comes up with if given a basic harness to render and iterate. They’ve all gotten pretty good at basic one shot pelicans, but seeing what they do when given a direct goal, and even getting a time lapse of each image as they edit and adjust would be really fun.
"Pelican on bicycle in SVG" is an adversarial example on multiple levels - out of distribution, ill choose of tool for the task, and anatomical impossibility of real pelicans using typical bycicles.
A lot depends on what the models were specifically trained for using reinforcement learning. Modern models are generally OK and everything, but they're only really good at the things they have extensive RL for.
Yes. I've been doing a bunch of 3D experiments with recent models, you can check the session logs and see if they've retrieved anything. They're building the 3D models directly in code.
I even built an environment that allows execution of Python code to render with Blender but allows no other interactions (including networking, I sandbox that off). They can create good 3D models just fine - here's a Faberge Egg by Gemini 4 Argon: https://tools.simonwillison.net/blender-viewer?url=https%3A%...
The supervisor agent will keep the session in a loop until 6 hours have passed and eventually the agent will decide to use up the remaining time rather than fighting with it
I wonder how much co2 is being thrown into the atmosphere everyday through the steady stream of “look what I made this LLM do” and endless “benchmarking”?
Intuitively I'd say that just-for-fun AI projects emit orders of magnitude less CO2 than what just-for-fun road trips or air travel emit.
So unless you can show that it approaches all the other frivolous consumption that humans love to waste resources on, maybe we can get back to experimenting with cool technology and talking about it, on this website called "Hacker News"?
One useful hint here is the API token cost - in this case "about $74 in API tokens" for Opus 5.5.
We know Anthropic run inference at a margin, so that $74 means that the cost of the electricity involved is substantially less than $74. I'd love to know the actual cost there.
Approximately nothing - the carbon cost of electricity is already very low, and most of the inference costs are GPU and datacenter amortization. You could require all AI inference to be carbon-neutral and that'd barely raise the API prices.
I wonder how much co2 you use to get through the day, and how much you use to remark on other people's creations in a way that suggests you disapprove of their utilization of their available resources. How much co2 do your projects emit, since you seem to be interested in those metrics?
I strongly suspect one HN post (or two now) is not even in the same ballpark. I could ask Claude to figure it out but that would be a complete waste of time and energy. ;)
reply