flimflamm

I configure it by copy pasting the pages to chatgpt and asking questions. Dumb? Maybe but seems to work.

We humans have a really good system to do this in practice. We use _imagination_ to simulate how things work and ... then we understand it.

However, there was only 57 however's in the paper it self.

Cost/manufacturing complexity. If you are country struggling to defend your self you don't think problems in 30 years if today problem is does the country exists or not. Might be difficult to put your self to a small defending countries shoes which is absolute running our of resources.

If you don't have workflows which repeat in inet you don't need openClaw. - Messages from school where to react - Getting payments from someone and tracking that you get them - Summary of news the way you like it from sources you like it every day - Integrated task lists reminders - Drafting taxation reports based on spending etc etc.

I just don't trust "the claw" so I build following system - Docker 1: * Locked up Claw docker - user level priv. Access outside to "one port" only.

- Docker 2: * Tool gateway with pre-baked commands - openclaw can only index what command to execute * Keys are here * Telegram hook to approve all "post" commands i.e. sending email or posting something somewhere.

- Docker 3: * LLM gateway keeping track of cost and routing

No as you could have multiple examples of expected output in single prompt. You should just "ask" - that's zero shot. If you "ask + provide" examples then you are in the n+1 shot realm. But I suppose terminology is shifting from this...

What confused me is the fact that in the paper all logical steps are give. It basically check that when all relevant facts are provided explicitly as links , how far and how complex a chain can the model correctly follow before it breaks down?

So it's simpler than "reasoning". This is not necessarily a bad thing as it boils down the reasoning to a simpler, more controlled sub problem.

Is this just buttering the cost of switches by crippling the optimization options compiler have?

I would focus on lowest of the axis. It does not help if some of the axis are at 100% if one of the axis is lacking.

Really really nice. Thanks for this!

Are the rules versioned somehow? I didn't see that.

Good to note that OpenAI solved 12/12 and DeepMind 10/12.

I wonder when Elon will go to nuclear business as green values have gone down (solar tiles anyone?)

I wonder what process in ASML is such that Mistral group would bring something new there...

Depends on the country. In Finland, it's ok to record your own discussions. Whether the recorder is BK (a third party) or the cashier is an interesting question, though.

Creating a database of recordings without user being able to know/influence is clearly violation of GDPR IF there is PII. That's going to be costly for BK.

Hah they kind of found "NULL" pointer in LLMs.

Have you looked in to "PSD2 and Open Banking regulation in Europe" ?

Except if you are a Finnish person who lives right next to Sweden (SEK) and Norway (NOK) who are not using Euro (and Russia but that's a different story).

I wonder if 6 would be understanding own thinking. Currently humans don't understand this. Thoughts just pop in to our heads and we try to explain what caused them.

Surely you can have the same LLM review the code. But treat output of that process similar to linters.

That one tool (LLM) didn't show issues - good. Then lets roll slaves back and start really checking if the meaning is correct and is the code implementing the right things the right way.

Can't CTRL-Scroll to zoom to the generated image which is with so small text that I can't read it.

It would be interesting to see how far one could get with fine tuning and RL. One could for example take free scad models, give 2D pictures from different angles and ask LLM to recreate the 3D design in scad language. Then compare volumetrically how close they are and provide feedback.

Once that is done then ask LLM to create a prompt and compare outputs etc..

Seems like there are some inaccuracies when zooming in. I don't think that Mandelbrot contains this kind of smooth features https://mandelbrot.site/?re=-0.11582128424197435&im=-0.88580...

Not illegal in Finland at least.

Imagine your eye's effective integration time (the period over which it "samples" light) shrinks during a rapid saccade—say down to about 5–10 ms. Under steady conditions, our flicker fusion threshold is around 60 Hz, but that’s because our retina integrates over a longer window.

If we want to “smooth out” the PWM cycles so we don’t see discrete pulses, we need a few cycles (say, 3–5) within that 5–10 ms window. In other words:

    For a 10 ms integration window:
        3 cycles → f≥30.01 s=300 Hzf≥0.01s3 =300Hz
        5 cycles → f≥50.01 s=500 Hzf≥0.01s5 =500Hz
    For a 5 ms window:
        3 cycles → f≥30.005 s=600 Hzf≥0.005s3 =600Hz
        5 cycles → f≥50.005 s=1000 Hzf≥0.005s5 =1000Hz

And yes, what we are now seeing in cars is super annoying. Similar disregardful to user comfort can be seen with HUD displays (at least with Volvo).

Paper also here https://arxiv.org/pdf/2409.13731

"That's not how it's described" - Thanks for the correction!

To create a patch, a small model is used to predict the likelihood for the next character in the input string. Input string: 'Lazy dog jumped over a fence.' Use the model to predict the likelihood of each character.

For example:

    100% sure the next character is 'a'.
    Or maybe it's 10% sure it's 'a', 10% sure it's 'b', and so on.