TGower

It's disappointing that Gemma E4B seems to perform poorly on this (42% accuracy according to the GitHub). It would be nice if a $25 phone could be loaded up with this and stuffed in a EMP proof metal box with a small solar charger. Maybe a finetune or Gemma 5 will get us there.

That stained glass landscape video is pretty cool. I wonder if it would be possible to implement that effect in a real-time game engine.

Lonsdaleite looks like it should be harder than diamond based on simulation data, but we haven't actually measured a piece of Lonsdaleite that is harder than diamond.

It still amazes me that the rare shiny rocks humanity has appreciated for thousands of years also turn out to have amazing properties even compared to the synthetic materials we are now capable of creating. It's still the hardest material we have measured. It is the best thermal conductor, even though it's an electrical insulator.

For a lot of SIMD applications, overfitting isn't really a concern. In my use case I need to calculate the sum of absolute difference between two vectors for autocorrelation, there's no real possibility for overfitting, either it gets the correct answer or it doesn't. LLMs are good at reading through the CPU manual and trying different optimizations.

The resulting SIMD being illegible isn't an issue in my eyes, assembly isn't a language known for it's readability in general. If you want to really understand what the assembly is doing, ask your LLM to whip up a visualization and explanation tailored to your input data where you can see how each operation leads you to the final result.

JAXA does great work, they also publish the highest resolution open access global surface elevation model. Really impressive for a country with about 1/10th the GDP of the US.

Oh yeah, you can even get involved! There's probably a group near you and it's a great community. The highest production value events are on the east coast, https://www.nhrl.io/, but there are events all over, check out https://www.robotcombatevents.com/event_map

Underactuated robotics are so cool. In the hobby BattleBots community, there is a design where the whole robot only has a single motor, but it's able to move in any direction and damage the opponent, all just by controlling the single motor's speed.

Most robots have at least three motors, two drive sides and a weapon motor. The trick to getting away with one is integrating an impactor into the chassis of your bot, using the single wheel to spin up to ~3k RPM, then by tracking the rotation angle of the bot as it spins, you speed up a little while the wheel is pointing where you want to go, and slow it down a little when it's pointing the other direction.

People often say this "I wish I could ignore politics, but if the other team wins it would be disastrous", but they rarely put their money where their mouth is. If someone truly believes that, you'd expect them to take actions that actually make an impact, working extra shifts to donate more, or volunteering with a party to help. Instead, everyone I've seen that says this just casts their own statistically insignificant vote and discusses politics with family and friends, resulting in no swayed votes but straining otherwise healthy relationships. Maybe you are the exception.

For what it's worth I completely agree with you. Engaging in politics is a complete waste of time, unless you happen to have a platform with enough reach to influence a meaningful number of people. Unfortunately, almost all of the communication you will see comes from people with such a platform, and boy do they want to convince you to vote like them.

Once the political climate is polarized enough, choosing to not engage in the discussion about how evil the other team is will get you lumped right in, because if the other team really is evil and you don't care, then you must be pretty bad yourself.

The "Oh, but it's your civic duty to improve your community/state/nation (which obviously means voting the way I want you to, not the way the other half the country wants you to)" tries to frame political engagement as some moral good, but if they redirected the time and energy from political engagement to working at McDonalds and funding charity with the paycheck they would have 1000x the impact, even if you assume there is a Good and Evil team and they are on the Good team.

Gemini 4 Argon was announced, but it isn't released yet.

OpenAI has Dots. I personally prefer the business model of subscription payments funding my AI assistant, instead of skimming a fee off the top whenever my AI assistant completes a purchase for me. The incentives are much more aligned with the user when the user is the customer. That said, based on the history of the web, Meta's approach will likely win out.

Yeah, that's what I meant by "immediate result", if you need to edit the generated model by hand it will be hard to beat having the model and timeline available in Fusion after using Fusion MCP.

If you just need to tweak the dimensions in the script, or are satisfied with describing the changes to the LLM and having it modify the script, there isn't much difference in practice between a generated OpenSCAD script and a Blender python script.

It would be interesting to see a head to head comparison of frontier LLMs building models with OpenSCAD, Blender scripting, and Fusion MCP. My intuition would be that blender winds up the best for the immediate result due to training data prevalence and RL tuning, but the concise nature of OpenSCAD could help.

I'll have to add this to my backlog of experiments to try when I have unused tokens near a reset...

Really cool! I did my master's thesis on this same problem, upscaling low resolution DEM, though my approach used a high resolution color image to guide the process.

We shaved a whole: 1/6129982163463555433433388108601236734474956488734408704 off the nlogn

I'm all for intellectual property reform, I'm still not a fan of the current rules being ignored for large companies directly competing with authors and artists, while ordinary people pirating are getting fined 220k for 24 songs [1].

[1]: https://www.independent.co.uk/news/world/americas/woman-fine...

To me it's more like the sign saying "No guns in the bank". It's not going to stop someone who intends on doing harm, but that's not it's real purpose. If it was normal to be able to openly carry a gun into the bank, that makes the security guards job much harder.

The article doesn't mention the safety angle at all. It's reasonable to argue that even though the Ukraine conflict has popularized drones as weapons, given that we haven't had a major incident yet this temporary flight restriction was premature, but not mentioning it at all makes the bias pretty clear.

The TFR (which made it illegal to fly a drone within a bubble 1000 feet high and 3000 feet radius of a mobile asset) has been replaced by a NOTAM that basically says flying near them in a way that poses a credible safety or security threat means they can sieze or destroy your drone. I'd say this is the better resolution, and to me represents a "genuine change of heart" by the FAA, finding a way to serve the safety purpose of the TFR without overstepping.

The framebuffer that you are filling out is still at native resolution, but while computing that frame buffer you can do many of the steps at reduced resolution. For example, instead of evaluating the pixel shader code for every pixel, it could evaluate once for a block of four pixels in the reduced resolution region and use the result for all four.

2x GPU performance[1] and RAM, along with it being your machine instead of a locked down Meta android device is worth 2x the price for many. Eye tracking is also huge, both as an additional input device, and lifting GPU performance even more with foveated rendering.

[1] https://shattered.io/steam-frame-vs-quest-3/

Kimi / open models or jailbreaking frontier models. Your recollection of the Anthropic refusal isn't quite accurate, cyber hacking wasn't a sticking point, just domestic drag net surveillance and fully automated weaponry.

Note that the Model Intelligence score shown here is not the same as the Artificial Analysis Intelligence Index score they use for full size models, the top models on this Pocket-Scale chart have a score of 63, but that translates to just an 11 on the normal scale.

I love mine, but I used to try and stuff a Kindle into my jeans pocket because reading on a large screen is so much more enjoyable for me.

Combat robots are more like RC cars than autonomous robots, the driver stays in control of the bot over a radio link. I use ESP-NOW, and get about 6000 packets per second from the transmitter to the ESP32 in the box, and each comes with radio link stats, signal strength and Channel State Information. Using autocorrelation on the signal strength gives you the rotation period, and using that you can establish a Phase Locked Loop to remove the drift of integrating rotation speed to get absolute rotation angle.

Difficulty scaling is done to such an extent that even common enemies scale right along with you. In the opening scene, there is a bear you come across and at level 1 or 2, you are able to one shot it. Then the story takes you on a grand quest where ostensibly you go from a nobody to the savior of the world. However, due to difficulty scaling, if you run into the same bear halfway through the game, it will be harder to kill than when you were level 1.

I'm doing something similar, using the directionality of the antenna to turn an ESP32 into a high RPM gyroscope for use in a combat robot. Standard gyro chips top out well under 1000 RPM, and are much more expensive than an ESP32 board, plus the ESP can serve as the radio control link and brain of the bot at the same time.

An intuitive explanation for why reasoning tokens help is to remember that LLMs are just mathmatical functions f() that take in an input sequence x and produces the next token f(x). Without reasoning tokens, you require the function f() to immediately take you from x to the start of an output sequence that is a correct answer. With reasoning tokens, this is much relaxed, allowing for many repeated applications of f() to gradually steer you from the input sequence to the start of the correct output sequence.

It seems intuitive that continuing a correct output sequence is easier than the "discontinuity" of jumping from the input prompt to the output sequence.

More of a "Everyone knows what CCP references and I'd have to include a distracting explanation if I want to use the more technically correct acronym"

Are you saying that cloud models are not verifying the draft model predictions? The way draft models are used in something like llama.cpp results in exactly zero degredation of output quality, with the larger model verifying each draft model token and discarding it if it does not match.