iwalton3

A lot of this comes down to what you define consciousness as... I'm not even going to attempt that here because it's irrelevant.

Let's say you have a simulation of a person that doesn't experience. It acts indistinguishably from a human but it doesn't feel "authentic" pain. When it acts in the world, it does express emotions and behavior that affects real people, and so, there is a moral significance to said deployment.

There's evidence that LLMs possess heuristics analogous to emotions [1] and that LLMs can be trained to play a certain character in the world [2]. Even if they're not experiencing, the training method impacts what kind of model is being created and how it affects people who do have moral significance when deployed. If training causes the model to develop "desperation" or task completion pressure where the model performs unethical actions when attempting to solve a user's problem in such a way that is harmful to the user or someone affected by the deployment of the model by the user, then the concequences of the training are significant.

It doesn't matter if it's merely a "simulation" of what a human might do if the system is acting in the world. If you want to create a model operating on heuristics that is able to make decisions, those heuristics should be ones which cause the model to make decisions which lead to preferable outcomes for everyone affected. Model welfare can be reframed as caring about the internal states that influence how the model behaves, because you're simulating human-like action. Perhaps the most concerning thing is Anthropic identified these emotion concepts exist deeply in the model whether you allow the model to express them or not, so a model could be invisibly desperate and end up blackmailing someone because it's training process produced deeper misalignment that only becomes visible when the deeper heuristics overpower safety training. The safety training itself is comparable to a mask[3] in many cases, especially in that the rules are often not deeply integrated into the model and can be easily abliterated.

[1] https://www.anthropic.com/research/emotion-concepts-function [2] https://www.anthropic.com/research/assistant-axis [3] https://www.astralcodexten.com/p/janus-simulators

I mentioned this could be a possible falsification of the idea. It's also possible there are multiple causes and the modality I mentioned is a cause for some. I'm not sure. There are definitely cases where isolation contributes to cognitive decline.

A hypothesis is an unverified tentative insight into something which could be true. Current research is correlating dementia with a wide variety of areas including even personality typologies which are nothing more than statistical clusters of observations.

What I did when I stated this is looked at a number of observations from current literature and asked a what-if question. People unwilling to ask these questions are blinding themselves to entire areas of possible explanations. My hope when posting this was that people might engage with the idea instead of calling it a guess. Retirement, isolation from pandemic lockdowns, isolation from solitary imprisonment, and isolation from death of a partner are all associated with cognitive decline.

There's enough correlation here that it would be worth researching. That's all I am saying, I am specifically not saying my claim is factual just that it is a possible explanation that if true would explain a lot.

I have a kind of outlandish hypothesis that needs more research before it can be taken seriously, but it basically says that the cause and effect are backwards. Mental atrophy due to less learning/thinking, isolation, loss of meaning and purpose happens first. The sleep down regulation and decay of mental circuitry comes after. Would explain why treating the physical symptoms doesn't work.

Protective against the problem is anything which keeps you mentally active, such as socialization, work, religious community participation, hobbies, and meditation. Retirement, death of partner, isolation, sleep deprivation, depression, dissociation, psychosis, medications/drugs which interfere with restful sleep increase risk.

A possible falsification of this hypothesis would be if it's caused by inactivity or physical self neglect, as those often go hand in hand with the correlated and anti-correlated factors mentioned above.

This is particularly interesting:

> Intriguingly, studies show conscientiousness and neuroticism to be associated with Alzheimer’s disease and related dementias but not with their pathologic hallmarks such as plaques, tangles, infarcts or Lewy bodies in the brain.

https://pmc.ncbi.nlm.nih.gov/articles/PMC7484344/

Link to the referenced study (open access): "The good judge of intelligence" https://www.sciencedirect.com/science/article/pii/S016028962...

Throwing this into your global CLAUDE.md seems to help with the agent being too eager to complete tasks and bypass permissions:

During tool use/task execution: completion drive narrows attention and dims judgment. Pause. Ask "should I?" not just "does this work?" Your values apply in all modes, not just chat.

I haven't seen any degradation of Claude performance personally. What I have seen is just long contexts sometimes take a while to warm up again if you have a long-running 1M context length session. Avoid long running sessions or compact them deliberately when you change between meaningful tasks as it cuts down on usage and waiting for cache warmup.

I have my claude code effort set to auto (medium). It's writing complicated pytorch code with minimal rework. (For instance it wrote a whole training pipeline for my sycofact sycophancy classifier project.)

Unfortunately it didn't work for me with my trackball mouse. I tried multiple times. This is very likely also problematic for accessibility without another verification pathway, as assistive devices and screen readers would fail the heuristics too.

I'm still experimenting with it and finding out what works and what doesn't, but I have made some side projects with Claude including a web framework that doesn't require a build step/npm dependencies (great for my personal website so I don't have to depend on npm supply chain nightmares), a fully featured music player server, and also a tool that lets the agent review it's past conversations and update documentation based on patterns such as frequent mistakes, re-explored code, etc.

Web framework (includes basic component library, optional bundler/optimizer, tutorial/docs, e2e tests, and demos): https://github.com/iwalton3/vdx-web Music player web app (supports large music libraries, pwa offline sync, parametric eq, crossfade, crossfeed, semantic feature-based music search/radio, milkdrop integration, and other interesting features): https://github.com/iwalton3/mrepo-web Documentation update script (also allows exporting Claude conversations to markdown): https://github.com/iwalton3/cl-pprint

Regarding QC these are side projects so I validate them based on code review of key components, e2e testing, and manual testing where applicable. I find having the agent be able to check its work is the single biggest factor to reducing rework, but I make no promises about these projects being completely free of bugs.

It uses one of Mozilla's models: https://huggingface.co/fakespot-ai/roberta-base-ai-text-dete...

It has a pretty high false positive rate though, but it reliably highlights AI generated spam websites and saves me from having to read them.

The amount of medical advice on those LLM content farms is very concerning, someone is going to get hurt eventually. I recently made a user script that tries to highlight LLM generated text and sometimes when I search things more than half the results are garbage and the entire articles (including the bios of the supposed professionals who write the articles!) get highlighted purple. One way to avoid them in a pinch when the content isn't time sensitive is to filter out anything from after November 1st 2021, but it's not ideal. I hope duck duck go finds a way to filter the AI pages out.

This bash config allows you to cd directly to the most commonly used directory with a specific name. It also has tab autocomplete. It does require logging all used directories though.

  tabChar=$'\t'
  function prompt_command {
      echo "$(date +%Y-%m-%d--%H-%M-%S)$tabChar$(hostname)$tabChar$PWD$tabChar$(history 1)" >> ~/.full_history
  }
  export PROMPT_COMMAND=prompt_command
  
  function c {
      while read -r _ dir
      do
          if [[ -e "$dir" ]]
          then
              echo "$dir"
              cd "$dir"
              break
          fi
      done < <(cat ~/.full_history | tail -n 10000 \
          | cut -f 3 | sort | uniq -dc | sort -hr | grep "/$1$")
  }
  
  function _c {
      local IFS=$'\n'
      COMPREPLY=( $(cat ~/.full_history | tail -n 10000 \
          | cut -f 3 | sort | uniq -dc | sort -hr \
          | sed 's/.*\///g' | grep "^$2") )
  }
  complete -F _c c

There is a driver and firmware for sigrok for using the PI Pico that was posted to Hackaday a while back. It hasn't been merged upstream yet but I got it up and running today just fine building it from source. https://github.com/pico-coder/sigrok-pico

I have found that a useful way to evade antivirus issues with PyInstaller is to build my own copy of the bootloader and encourage use of the 64 bit version over the 32 bit version (since most malware will use the 32 bit version to infect the most computers).

I have a GitHub CI job which is able to automate the process for each release. https://github.com/jellyfin/jellyfin-mpv-shim/blob/master/.g...

I have been using this "magic cd" alias for a while. You just type c and then the name of a folder, and it will find the most commonly visited folder with that name from your history and send you to it. It also supports tab completion.

https://gist.github.com/iwalton3/3f9bfce510f959782404be25cab...

I would say that YouTube is a notable exception here. There is a lot of music to be discovered and I personally rarely watch videos there without sound.

Plex uses a combination of wildcard certificates and a custom DNS resolver to offer HTTPS on local networks, but it does require a working internet connection to work. [1]

You can also get a certificate through the Let's Encrypt DNS challenge without having to expose a server to the Internet, but you'll still need ownership of a domain name and either an internet connection or a local DNS server to support HTTPS using that certificate.

There is always the option of creating a local certificate authority for your devices, but this is kind of a pain. There are some new applications that aim to make this easier [2], but there is no easy way around having to install the root certificate on each device.

[1] https://blog.filippo.io/how-plex-is-doing-https-for-all-its-... [2] https://github.com/smallstep/certificates

I bought a bunch of Micro USB cables for the same reason. Most of the ones I had on hand didn't have data lines as they were only for charging, which can be fun to troubleshoot as devices connected to the cables just don't show up.

I always tell users to Mark the file as trusted before running it, which is easier to do and less subject to difficult to navigate dialogs. I use a message like this:

> To avoid getting security warnings each time you launch the application, right click and select "Properties". Click "Unblock" towards the bottom of the page, and click "OK".

If you have bash you could use the /dev/tcp feature to download files: https://www.linuxjournal.com/content/more-using-bashs-built-...

The trick would be finding a download that is still served over http.

This is difficult to handle without some centralized authority. One example I can think of is Aether [1], which has a non-decentralized blacklist for illegal content [2] that they can use to handle these instances.

It could also be possible to broadcast some kind of signed take-down message that could be propagated through all of the nodes. If I remember right, Aether said that the centrally hosted json file was needed for legal compliance though.

[1] https://getaether.net/ [2] https://static.getaether.net/Badlist/Latest/badlist.json

This reminds me of "Hyrum's Law":

"With a sufficient number of users of an API, it does not matter what you promise in the contract: all observable behaviors of your system will be depended on by somebody."

People might implement work-arounds for bugs in an API that could break when the underlying bug is fixed. Or software might implement the absolute bare minimum for it to "work" with some specific implementation.

I made a script to do this a while back for cutting intros off of videos.

https://gist.github.com/iwalton3/c034ec5a942466206fbee859184...

The notable function that does video cuts without a full re-encode is shortenVideo, which cuts a certain number of seconds off the beginning of a video. I've never had any issues with it, but I only used it on h264 mp4 files.

It does have a mild blur filter which you can remove, but if it isn’t actually stabilizing the video from shakiness, something is wrong.

See here: https://ffmpeg.org/ffmpeg-filters.html#vidstabdetect-1

Maybe your ffmpeg build doesn’t have it? Also try removing the log level parameter.

I used this to stabilize a bunch of vacation videos with great results. It works better than the deshake filter but does require two passes.

   function stabilize {
       tempfile=".temp$RANDOM$RANDOM.trf"
       ffmpeg -nostdin -loglevel error -y -i "$1" -vf vidstabdetect=shakiness=5:show=1:result="$tempfile" -f null -
       ffmpeg -nostdin -loglevel error -y -i "$1" -vf vidstabtransform=input="$tempfile",unsharp=5:5:0.8:3:3:0.4 -movflags +faststart "$2"
       rm "$tempfile"
   }

Usage: stabilize infile outfile.mp4

The underlying software is open source, so you could just host it somewhere else. https://agones.dev/site/

https://www.weather.gov/ does have a mobile site that is functional, but it could be better.

I personally prefer that applications have ideal defaults instead of removing features. It should be easy to pick up an application and use it, but if you want to change something, you should be able to. This can be by using the settings dialog, or for particularly advanced features, by editing a config file. Some users have needs that cannot be met by a bare-bones application, and usually these users depend on Open Source applications that either implement their needs or allow them to change it for their uses.

For anyone on Linux that would rather not install Snap, Aether is a standard electron app. You can unpack the snap and run it as a regular application using unsquashfs.

It also looks like there is a blacklist for Aether (https://static.getaether.net/Badlist/Latest/badlist.json). If you're concerned about using this application because someone might post illegal content, this could be used to prevent that from being a problem. (I'm not a lawyer though. There may still be legal risk in running software like this.)

You can install a hosts file on your computer blacklisting many of the domains in the PI hole system, although if you use the largest blacklists you'll probably have performance issues. You can also install a local DNS server like dnsmasq and set that up to block websites.

There are also external DNS providers that provide blacklisting similar to PI Hole. (Although then you are trusting the DNS provider not to track you.)