Fable was really excellent before the whole fiasco with the US government. Once they brought it back it was not the same at all. I have switched over to ChatGPT now for most things, its answers are way better than Fable in my experience. I still use Opus 5 for purely coding tasks.
This is why I think open weight models will win out in the end. Right now there’s too much going on behind the scenes with the models. Day to day you never know if you’re going to get smart Claude or dumb Claude.
that part is extremely frustrating, I can't tell if it's a placebo or if the model quality does actually vary. the uncertainty makes me more hesitant to rely on it heavily as some days it just seems incredibly dumb to the point of being useless.
I agree, I wish more work was being done on building a hydrogen economy. Hydrogen can be generated sustainably, is a great way to store excess energy from renewable sources, can be transported, and as a bonus can be mixed with hydrocarbon fuels to be used directly as a heating source.
As an example, a house in NH can generate a lot of excess solar power in the summer months when it uses hardly any electricity. If all that energy could be stored as hydrogen, in the winter (when solar power is less due to the far northern latitude and snow covering the panels) it could draw down on the hydrogen to generate electricity and also use it for a backup heat fuel.
I'm in Northern Europe and am in that situation. I have an electric heat pump, so would need around 3 MWh over the winter period. Apparently that would need a tank around 4m3 at 700 bar (typical pressure of hydrogen vehicle tanks). Plus the fuel cell actually generate heat, so that could be used for heating too. Doesn't seem terribly unrealistic.
I'm skeptical of any application for hydrogen where electricity could be used directly. In your case, I think hydrogen would better be used in a more centralized power plant as a backup for renewable sources (perhaps with the waste heat then used for district heating), and the power used to drive your heat pump.
The issue is long term, high capacity storage. To store enough energy to last an entire winter using batteries just isn’t practical today. I think you’d need around 100 power walls at today’s capacity to make a house completely off grid, meaning you store excess solar from the summer and draw it down in the winter. With hydrogen stored in pressurized tanks, a 20 m3 tank could be enough to last the whole winter, which seems much more practical than 100 powerwalls.
I think this could be a solution if you wanted a fully off grid, solar powered house, or perhaps if we collectively decide to generate more electricity at home and not rely so much on the grid.
Storing hydrogen in big underground salt caverns at reasonable pressures and converting it back into electricity in a big power plant is probably more efficient and safer than storing it in millions of tanks in cellars at 700 bar and using small fuel cells to turn it back into electricity.
Oh sure, batteries are completely inappropriate for seasonal storage. But for a home I think a synthetic liquid fuel would be better than hydrogen. Seasonal storage at grid scale is rather different; there I suspect thermal storage will win, if it is brutally cost optimized (so-called sand batteries don't qualify; their capex is too high.) Unfortunately thermal seasonal storage for an individual home isn't very practical; it could be for a village.
The biggest gain you'll get is faster memory, provided you have enough capacity to load all the weight into vram. The DGX sparks and Apple silicon memory bandwidth (and also memory access latency) drag down the decode speed quite a bit.
I have two GPU rigs both with 2x RTX Pro 6000, can get ~250 tk/s decode with deepseek-v4-flash in native mixed precision. For context, in antirez's dwarfstar project he only gets ~20-40 tk/s on the same model @ 2bpw on M5 Max.
The latter is for sure usable if it's your only option, but it's really hard for me to personally go back to speeds like that when I've experienced the former.
(Also worth noting dwarfstar only has experimental support for dspark spec dec, when that lands it will definitely give a big boost at higher acceptance rates)
That’s very interesting, thank you for sharing. Looks like it could be a very useful tool for testing high performance networking.
I wonder if something similar could be done using TC BPF instead of AF_XDP? My only reservation about AF XDP is that it requires a special NIC to support it, so it may not be useful for a “regular Joe” user. I wonder if TC BPF would also work since it similarly bypasses the Kernel networking stack, I believe you can put packets directly into the NIC TX queue for transmission
You can, but the interesting thing about AF_XDP is that you've got a userland path to writing directly to the card's DMA buffers; TC BPF still allocates an skbuff for every packet you send.
Also with liburing you have zero-copy send and receive operations for normal protocols like UDP or TCP, but this requires a NIC that supports scatter/gather DMA (so that the packet headers go to/from kernel buffers, while the data goes to/from userland buffers).
AF_XDP is also available in older kernels, but with recent enough kernels (zero-copy receive is a recent addition) and with a good NIC, liburing should provide a similar performance.
All of my childhood friends who bragged about being good at GH/RB only ever bragged about being good at the _games_. Never ran in to someone trying to pass it off as being able to play the guitar.
Anecdata, friend. Two old friends became extraordinarily good at guitar hero and even won competitions. They're also some of the best guitar players I know. One of them is an extremely good drummer and can do improv piano like a beast. They'd both agree that guitar hero helped them play better.
What are your thoughts on games like Keyboard Hero?
Maybe not, but folks keep telling me that I'm ignorant for not agreeing that being able to "engineer" a prompt is the same exact thing as being a poet or painter.
Another nice benefit of this would be location tracking. Once you had established the database of tools and the CV setup to recognize them, you could ask the LLM, “where did I last place my Stanley knife”?
Exactly. The challenge isn’t getting the LLMs to make sure they validate their own code. It’s getting the LLMs to write the correct code in the first place. Adding more and more LLM-generated test code just obfuscates the LLM code even further. I have seen some really wild things where LLM jumps through hoops to get tests to pass, even when they actually should be failing because the logic is wrong.
The core of the issue is that LLMs are sycophants, they want to make the user happy above all. The most important thing is to make sure what you are asking the LLM to do is correct from the beginning. I’ve found the highest value activity is the in the planning phase.
When I have gotten good results with Claude Code, it’s because I spent a lot of time working with it to generate a detailed plan of what I wanted to build. Then by the time it got to the coding step, actually writing the code is trivial because the details have all been worked out in the plan.
It’s probably not a coincidence that when I have worked in safety critical software (DO-178), the process looks very similar. By the time you write a line of code, the requirements for that line have been so thoroughly vetted that writing the code feels like an afterthought.
I'm becoming convinced that test pass rate is not a great indicator of model quality - instead we have to look at agent behavior beyond the test gate, such as how aligned is it with human intent, and does it follow the repo's coding standards.
also +1 on placing heavy emphasis on the plan. if you have a good plan, then the code becomes trivial. I have started doing a 70/30 or even 80/20 split of time spent on plan / time implementing & reviewing
Well, the water analogy actually holds up quite well if you consider the charge field moving, not the electrons themselves. This guy has a lot of great videos using water channels to explain electricity. It is fascinating how under a high speed scope, you can see the electrical Charge “flow” like water down each branch of a circuit.
Even rubber dams stretched across pipes work exactly like capacitors. Firstly, they block DC: water cannot flow. But back and forth movement is conveyed (AC passes). Less capacitance in series, more in parallel. Two such dams in series do not have more capacitance because to get one to stretch, the other must stretch; they partially cancel. And since there are two, more pressure is needed to get the same stretch. (Same as more voltage needed to cram in the same charge: less capacitance).
Inductance doesn't have an analog. To some extent, the inertia of the fluid cam model some of it, I suppose. Like what is "water hammer" in plumbing? The circuit is too suddenly broken, but the water wants to keep moving. There's gotta be a resulting momentary pressure rise there in the closed-off line, similar to voltage rising in an interrupted inductor. If the valve were some weak piece of crap relative to the mass of the water, the water would break it: like arcing.
Or just a long pipe where the inertia of the water resists change in motion.
This is what causes the "water hammer" effect which is a problem for plumbers, but a great thing for all kinds of fun experiments, e.g. creating predictable cavitation [0].
This was the stunning one to me. Just blew my mind literally observing it run up a dead end wire and slosh and rebound like water.
Previously I was used to thinking of dead ends as simply functionally inert. That without a circuit, nothing at all happens in the dead end wire other than the potential for something to happen.
Sure I know something more than nothing actually happens since there is an elevated charge there. But still just the mental model shorthand is that no circuit = no nuthin.
But it's not. It's actually like a pipe with a little bit of air to allow for some compression, and even the dead end has a small flow that travels to the end and builds up against it, then rebounds back and eventially levels out at some homogenious but now higher pressure.
You don't screw up something this major, it doesn't happen by accident nor by incompetence.
They had plans to bomb something south of that airport, they had to postpone those plans now that the info is public enough that whatever their target was is definitely aware of those plans.
This is exactly what happened and not to be immodest but it was my first guess before it was confirmed. The closure was a miscommunication between the FAA and Pentagon set off by a balloon. This was pure incompetence and arrogance. This HN thread is almost unbelievable how many wild guesses were made.
They have confirmed it was for testing a counter-drone weapon. They did not say why they set it to expire in 10 days, that part seems like it was probably a mistake.
They gave you a plausible-enough reason and you took their word for it. That's completely fine, you are well within your rights to decide for yourself whether you believe them or not.
I don't, and since neither of us can know for sure given the info we've been given, it's useless for us to argue about our opinions.
I am curious, what explanation would justify a full closure of the airspace over a major us city for 10 days, in your opinion? That is the real screwup here. Whatever justification they are giving is entirely beside the point. Closing the airspace, even to emergency medevac flights, is negligence.
Miss me with your jUsT cUrIoUs, I have no need to make up hypothetical scenarios under which this would be justifiable. The burden of proof is not on me.
If there was ever a time when the old Soviet Union could have won the Cold War... Fortunately for us, the window of top-down incompetence came far too late.
Quite literally, the previous post on this blog is from 2024 talking about what a revolution the Rabbit R1 is. We all know how that turned out. This is why I give every new trendy developer tool a few months to see if it’s really a good thing or just hype.
Maybe that's why these users go crazy over openclaw, they may need or yearn for such a tool. I don't but that doesn't mean there isn't a market for it though.
There isn’t a market. OP wrote that Rabbit R1 post after seeing the release video (according to a comment on this link, their blog post says otherwise) and immediately called it a ”milestone in the evolution of our digital organ”. Their judgement is obviously nonexistent.
Something tells me they never even downloaded OpenClaw before writing this blog post. It’s probably an aspirational vision board type post their life coach told them to write because they kept talking about OepnClaw during their sessions, and the life coach got tired of their BS.
This is why I think open weight models will win out in the end. Right now there’s too much going on behind the scenes with the models. Day to day you never know if you’re going to get smart Claude or dumb Claude.