That just implies LLMs are suggestible. The same is true of children. As we get older and build a more complete world model in our heads, it's harder to get us to believe things which go against that model.

Tell a 5-yr old about Santa, and they will believe it sincerely. Do the same with a 30-year old immigrant who has never heard of Santa, and I suspect you'll have a harder time.

That's not because the 5-year old is dumber, but just because their life-experience ("training data") is much more limited.

Even so, trying to convince a modern LLM of something ridiculous is getting harder. I invite you to try telling ChatGPT or Gemini that the president died a week ago and was replaced by a body-double facsimile until January 2027, so that Vance can have a full term. I suspect you'll have significant difficulty.

> Do the same with a 30-year old immigrant who has never heard of Santa, and I suspect you'll have a harder time.

There's a plethora of people who convert to religion at an older age, and that seems far more far fetched than Santa.

> There's a plethora of people who convert to religion at an older age, and that seems far more far fetched than Santa.

Being in a religion doesn’t imply belief in deities; it only implies people want social connection. This is clearly visible in global religion statistics; there are countries where the majority of people identify as belonging to a religion, and at the same time only a small minority state they believe in a “God”. Norway is a decent example that I bumped into just yesterday. https://en.wikipedia.org/wiki/Religion_in_Norway

Sure.

But I bet you'd have a significantly easier time converting a child rather than a 30/40/50-yr old to a religion.

My point is that LLMs are suggestible, perhaps more so than the average adult, but less so than I child I suspect. I don't think suggestibility really solves the problem of whether something has AGI or not. To me, on the contrary, it seems like to be intelligent and adaptable you need to be able to modify your world model. How easily you are fooled is a function of how mature / data-rich your existing world model is.

I've gotta say, this writeup gives me an itchy feeling. It really does feel like poking around a synthetic brain at this point.

You could make the argument it's closer to the blocks of a CPU compared with a brain, and it's no different to copy-pasting some IP block for eg, HW JPEG decoding. But I feel like the difference here is we're 'discovering' these blocks / organs. They weren't designed, they were evolved.

At some point I will clean up and share the dynamic layer modification code for oobabooga Text-Generation-WebuUI.

You can enter the setting, and apply new re-layering architectures. Its very weird chatting with these brain-damaged models.

The difference is less stark these days, with generative design being used for semiconductors.

Altering these features isn’t messing with evolution anymore than tweaking a CAD file that used genetic algorithms: it’s all math, 1s and 0s.

Linux can fool you into that sense of security for a long time. But there will come a point where the facade crashes down.

Maybe it's plugging your laptop into an external projector, or getting to sleep and wake correctly without the WiFi driver segfaulting, or maybe it's trying to get HDR working, or audio routing or...

Or not. I've been using Linux since 2006, initially for fun and giggles, and then by choice, and nowadays it's without a doubt the superior alternative. My successive employers forcing me through Windows gave me the "privilege" of daily-driving Windows at work and seeing up-close everything that Microsoft has produced since XP. That show-stopping 15min drivers installation because you plugged-in some headset or mouse? That's Windows-only. That screen positioning that can't be remembered consistently between work and office? Ditto. That nefarious hours-long unskippable OS-upgrade? Same. That feeling of losing your mind because you configured something in a specific way and now it's gone? Oh yeah, that's Windows overriding your user-preferences behind your back, just because. Your playlist blasting in the open space? Apparently that's a Windows feature of your bluetooth headphones switching off (even when they were not playing because your playlist was paused). That wifi that you have to re-enter the credentials once a month or so? Yeah, Windows suspend/resume once again crapped and that somehow caused network amnesia. And that's just the hardware-ish stuff. The permanent nagging, upselling, silly distractions, adverts and overall lack of polish and stability is nothing new in software-land.

I cannot remember when I last had a network, bluetooth, audio or driver issue with Linux on my ThinkPad. I legitimately cannot. But I can assure you I that had 3 of those in the last couple weeks alone on my work's HP elitebook with W11 24H2.

> Linux can fool you into that sense of security for a long time. But there will come a point where the facade crashes down.

All my laptops through multiple jobs have been running Linux starting in 1994 (way before the "Year of the Linux desktop") with zero problems. I switched to Mac laptops in 2012 only because that's what work at the time gave me. In later years accumulated many Mac laptops but quality has been goind down fast. Next time around I'm back to Linux because can't take it anymore.

The author explicitly shows how in Apple's walled garden, even that had hurdles that have been put up, 'for security'.

If you're benchmarking something, old & well-characterized / understood often beats new & un-characterized.

Sure, there may be shortcomings, but they're well understood. The closer you get to the cutting edge, the less characterization data you get to rely on. You need to be able to trust & understand your measurement tool for the results to be meaningful.

Aren't human coders also nondeterministic?

Assigning different agents to have different focuses has worked for me. Especially when you task a code reviewer agent with the goal of critically examining the code. The results will normally be much better than asking the coder agent who will assure you it's "fully tested and production ready"

Human coders are far more reliable. The only downside is speed, and therefore cost

As someone who worked on kindle power consumption many years ago: One of the (by far) biggest consumers of power is the WiFi connection. It has to wake up and respond to the AP in order to not get disconnected every x seconds.

Off the top of my head, I think 'on' average power consumption was ~700uA without wifi, and about 1.5mA+/- with Wifi. This is from over a decade ago, so my memory is fuzzy though...

Obviously, page changes used relatively large amounts of power. I don't recall the exact amounts, but it was 100s of mA for seconds.

There is also an "every x pages, do a full screen refresh (black to white)" to fix up the ghosting issue that the article writer saw.

I removed the battery on mine, kept the battery chip and fed 5V into the battery terminals, from Kindle's USB connector, through a diode (so 4.4V-ish). Without a battery it needs something that can deliver at least 1.5A, for short bursts. An older powered usb hub seems to work fine, hub is connected to my raspberry pi, and I use ssh through usb networking, no wifi, no battery, worked fine for months now.

Awesome tips. I'll try increasing the refresh interval to 2 minutes and turning off the wifi in between refreshes to see if helps with battery life.

Side note this also finally explains to me why battery life on the Kindle is SO good in airplane mode.

Yeah you'll probably want to turn on airplane mode between the refreshes. That's at least what KUAL "Online screensaver extension" from Mobiread forums does, with it I'm getting several weeks of battery life with refreshes every couple of hours.

Could you adjust the refresh frequency based on your usage patterns? So refreshing less frequently outside your main transit times. An extension of your current pause at night.

I never connect my RM2 to wifi and it's crazy, I don't charge it for months. Granted I use it maybe under 30 time a month too. I guess WiFi isn't necessarily disabled but yeah.

Haven't modded my paper white kindle, I'm thinking at the very least I'm going to get rid of those forced ads you read when you wake it up.

wonder if small amorphous silicon photovoltaic cells glued around the kindle would provide enough power to not charge every 5 days

Same.

I've been exploring some computer vision recognition stuff. Being able to reason through my ideas with an LLM, and make visualizations like t-SNE to show how far apart a coke can and a bag of cheetos are in feature-space has been mind blowing. ("How much of a difference does tint make for recognition? Implement a slider that can show that can regenerate the 512-D features array and replot the chart")

It's helping me get an intuitive understanding 10x faster than I could reading a textbook.

No.

I'd argue that no one would really care given it's GCC.

But if you worked for GiantSodaCo on their secret recipe under NDA, then create a new soda company 15 years later that tastes suspiciously similar to GiantSodaCo, you'd probably have legal issues. It would be hard to argue that you weren't using proprietary knowledge in that case.

Given that GCC is not public domain, the copyright holders will probably care.

Two socks and two sets of googly eyes are required for this exercise. Attach googly eyes to socks. Insert hands in socks.

Now I'm imagining Musk in a k-hole just staring at his hands in sock puppets for hours.