More accurately, I think it's better to say "I need light, therefore I need the lamp to have a lightbulb."
Inanimate objects don't have needs. They just are (exist).
RE: https://fedi.reclaim.technology/objects/69fefe52-0d76-44ff-96a0-54afab309052
https://codeberg.org/bajsicki/gptel-got
New update. Long time coming.
This is an extension to gptel. Might work as an alternative to gptel-agent (I really didn't like that overlay UI - I like seeing exactly what the guessing machine is doing step by step).
Also includes rudimentary permission and privilege escalation mechanism for agents, customizable sub-agent spawning, basic Lisp image interaction via SLY, better org mode tools (I think?), as well as granular configuration options for controlling exactly what is retained in the LLM's context window.
Code quality not guaranteed in the slightest. Been poking at this over the last month little by little, to make my local assistant more useful.
#emacs #gptel #llm
I forgot where I found this, but it is hilarious.
Edit: this guy’s name is Adrian Bliss and he has a youtube channel. For more, see https://www.youtube.com/shorts/wChMzO7nciM be a doll and give him a like or subscribe
Are there any good #1password packages for #Emacs?
The few I have found are strangely difficult to get running.
Yeah, I'm not down with this. Mass exodus from GitHub when?
RE: https://front-end.social/users/heydon/statuses/114982836440577365
Man I really wish I had some kind of tool that translates Python code into Common Lisp, which would then 'just work'... then I could more easily set things up, since I could just connect to the image and make the necessary changes, instead of having to write a file, and then restart the process hoping that I haven't broken anything.
Common Lisp is powerful because of its interactivity - you can make a change, try to compile, and it'll not only throw an error, but also keep on running with the already-working code without a hitch.
I'm sure there's some jank workaround in Python that lets you connect to the process and do the same thing, but with Lisp, it's just built-in.
#commonlisp #work #pythonhate #python
gptel-org-tools update.
1. Cloned to https://codeberg.org/bajsicki/gptel-org-tools, and all future work will be happening on Codeberg.
2. Added gptel-org-tools-result-limit and a helper function for it. This sets a hard limit on the number of characters a tool can return. If it's over that, the LLM is prompted to be more specific in its query. Not applied to all tools, just the ones that are likely to blow up the context window.
3. Added docstrings for the functions called by the tools, so LLMs can look up their definitions.
4. Improved the precision of some tool descriptions so instructions are easier to follow.
5. Some minor improvements w/r/t function names and calls, logic, etc. Basic QA.
Now, as a user:
1. I'm finding it increasingly frustrating that Gemma 3 refuses to follow instructions. So here's a PSA: Gemma 3 doesn't respect the system prompt. It treats it just the same as any other user input.
2. Mistral 24B is a mixed bag. I'm not sure if it's my settings or something else, but it fairly consistently ends up looping; it'll call the same tool over and over again with the exact same arguments. This happens with other models as well, but not nearly as frequently.
3. Qwen 2.5 14B: pretty dang good, I'd say. The Cogito fine-tune is also surprisingly usable.
4. Prompting: I have found that a good, detailed system prompt tends to /somewhat/ improve results, especially if it contains clear directions on where to look for things related to specific topics. I'm still in the middle of writing one that's accurate to my Emacs set-up, but when I do finish it, it'll be in the repository as an example.
5. One issue that I still struggle with is that the LLMs don't take any time to process the user request. Often they'll find some relevant information in one file, and then decide that's enough and just refuse to look any further. Often devolving into traversing directories /as if/ they're looking for something... and they get stuck doing that without end.
It all boils down to the fact that LLMs aren't intelligent, so while I have a reasonable foundation for the data collection, the major focus is on creating guardrails, processes and inescapable sequences. These will (ideally) railroad LLMs into doing actual research and processing before they deliver a summary/ report based on the org-mode notes I have.
Tags:
#Emacs #gptel #codeberg #forgejo #orgmode #orgql #llm #informationmanagement #gptelorgtools
PS. Links should work now, apparently profile visibility affects repo visibility on Codeberg. I would not have expected that.
PPS. Deleted and reposted because of strong anti-bridge sentiment on my part. Screw Bluesky and bots that repost to it. Defederated: newsmast.*
gptel-org-tools update.
Edit: there's some kind of issue with @Codeberg@social.anoxinon.de which prevents the link from working (returns 404). The old (but up to date) repo is here: https://git.bajsicki.com/phil/gptel-org-tools
1. Cloned to https://codeberg.org/bajsicki/gptel-org-tools, and all future work will be happening on Codeberg.
2. Added gptel-org-tools-result-limit and a helper function for it. This sets a hard limit on the number of characters a tool can return. If it's over that, the LLM is prompted to be more specific in its query. Not applied to all tools, just the ones that are likely to blow up the context window.
3. Added docstrings for the functions called by the tools, so LLMs can look up their definitions.
4. Improved the precision of some tool descriptions so instructions are easier to follow.
5. Some minor improvements w/r/t function names and calls, logic, etc. Basic QA.
Now, as a user:
1. I'm finding it increasingly frustrating that Gemma 3 refuses to follow instructions. So here's a PSA: Gemma 3 doesn't respect the system prompt. It treats it just the same as any other user input.
2. Mistral 24B is a mixed bag. I'm not sure if it's my settings or something else, but it fairly consistently ends up looping; it'll call the same tool over and over again with the exact same arguments. This happens with other models as well, but not nearly as frequently.
3. Qwen 2.5 14B: pretty dang good, I'd say. The Cogito fine-tune is also surprisingly usable.
4. Prompting: I have found that a good, detailed system prompt tends to /somewhat/ improve results, especially if it contains clear directions on where to look for things related to specific topics. I'm still in the middle of writing one that's accurate to my Emacs set-up, but when I do finish it, it'll be in the repository as an example.
5. One issue that I still struggle with is that the LLMs don't take any time to process the user request. Often they'll find some relevant information in one file, and then decide that's enough and just refuse to look any further. Often devolving into traversing directories /as if/ they're looking for something... and they get stuck doing that without end.
It all boils down to the fact that LLMs aren't intelligent, so while I have a reasonable foundation for the data collection, the major focus is on creating guardrails, processes and inescapable sequences. These will (ideally) railroad LLMs into doing actual research and processing before they deliver a summary/ report based on the org-mode notes I have.
Tags:
#Emacs #gptel #codeberg #forgejo #orgmode #orgql #llm #ai #informationmanagement #gptelorgtools
So currently I'm hosting gptel-org-tools on my own forge...
https://git.bajsicki.com/phil/gptel-org-tools
I can't seem to find a good way for people to create issues.
I thought that creating issues via email was a thing with Forgejo, but it appears not.
So here's a poll. Despite my hatred of Microsoft, is it for the best to move the repo to GitHub?
Or is there some way to federate with GitHub users so I don't need to open public sign-ups on my git?
I'm not exactly clear on what the move is here.
#Emacs #gptel #gitforge #forgejo #github #microsoft #foss
Factors:
1. Accessibility. Not everyone has really fast (or stable) internet.
2. Environmental. There's no reason to use more computing power than necessary for the task at hand. It's wasteful. Very few people need the fancy features advanced text editors introduce.
3. Interoperability. Text files I write and send are readable *everywhere.* Try loading up Google Docs on a 1024x768 screen with a 256MB RAM Pentium 3. You'll be lucky if Google Docs even loads.
4. Privacy. A text file is easy to protect. GPG is the most straightforward. It remains small, and there's no way middle-men can read it. Google Docs? Google has root and they're not encrypted from them. So, good luck.
5. Account requirements. Text files require no accounts anywhere. All you need it an Internet connection and a DNS server that'll point your computer the right way. SaaS requires that you also have up-to-date software, a powerful computer, and that you register an account with them to access files shared with you.
6. Storage space. A text file takes kilobytes. A .docx file takes megabytes. My daily journal, which granted has some meta-data but is still plain text, is nearing on 580kb after three years of diligent, detailed journaling. I can't help but doubt that Word would even open a .docx file that large if formatted natively. (Thousands of headings, links, timestamps, etc.)
6. Feature-set. Plain text lets you do enough for 99% of all tasks. Yes, it's not as pretty, but within the bounds of putting characters into a file, you have complete freedom. Proprietary services, on the other hand, have a very very rich feature-set, most of which is irrelevant for 99% of users. The drawback of this is that every user is forced to load these rarely-used functions onto their own computer when the applications load up. That's wasteful, and likely cost the world hundreds of millions in unnecessary energy expenditure already.
TL;DR: Use plain text unless you absolutely positively can't help it. It's seriously better in every way.
#plaintext #emacs #txt #notepad #bloat #bloatware #saas #googledocs #msword #microsoftword #rant
RE: https://fed.bajsicki.com/notes/a6uy06mot0
Normalize sending people plain text files instead of making them log in to proprietary services to read your two pages of notes.
There's zero reason to use things like Word or Google Docs for purely textual, written word.
There's zero reason why I should register with a service and use up a thousand (or more!!!!) times more bandwidth than I would if I downloaded your 2 kilobyte text file.
A direct link. Give me a direct link. Send me an email. Magic-wormhole it. Whatever.
Just be aware of the cost to the alternatives.
The fonts loaded by Google Docs alone could carry books and books worth of text. The javascript they push on end users itself is freaking massive.
Stop being sheep, use notepad.
/Yells at clouds
/rant
Just a snippet of elisp, that I thought may save some time.
I'm sure there's better/ more elegant ways of doing this, but jank that works is jank that works.
https://bajsicki.com/blog/elisp-neat-tree/
#emacs #elisp
And just a day later, here's a git repo with more philosophy than good code.
I think the philosophy part is more important personally, the code we can fix later.
https://git.bajsicki.com/phil/gptel-org-tools
TL;DR: Emacs (and its ecosystem) makes the whole vectorization/ RAG/ training stuff entirely redundant for this application.
Yes, this code fails still... a bunch, especially given the current lack of guardrails. But the improvement I've seen in the last few days makes me cautiously believe that with enough safeguards and a motivating enough system prompt, an active assistant may be possible.
Image attached isn't nearly as good as I've seen, but it's an off-hand example that demonstrates it working.
The only thing missing for me to be happy with this is one of those organic, grass-fed LLM models whose existence isn't predicated on theft.
#emacs #orgmode #gptel #orgql
RE: https://fed.bajsicki.com/notes/a6jw3n155z
So... I strongly disagree with LLMs (mostly with the marketing and the training data issue), but I found a use-case for myself that they may actually be 'alright' at.
I organize my life in #Emacs #orgmode It's great.
But over the years, my notes and journals and everything have become so large, that I don't really have a grasp of all the bits and pieces that I have logged.
So I started using org-ql recently, which works great for a lot of cases, but not all.
Naturally, I wanted more consolidation between the results, and better filtering, as well as a more general, broad view of the topics I wanted to look up in my notes.
So I started writing some tooling for #gptel, to allow LLMs to call tools within Emacs, and leverage existing packages to do just that.
It's in its inception, and works only 20-25% of the time (because the LLM needs to write the queries in the first place), but it works reasonably well even with smaller models (Mistral Small 24B seems to do alright with 16k context, using llama.cpp).
In general:
- It kinda works, when it wants to.
- The main failure point at the moment is that the LLM isn't able to consistently produce proper syntax for org-ql queries.
- The context window sucks, because I have years of journals and some queries unexpectedly explode, leading to the model going stupid.
So far it's been able to:
- retrieve journal entries
- summarize them
- provide insights on habits (e.g. exercise, sleep quality, eating times)
- track specific people across my journal and summarize interactions, sentiment, important events
It doesn't sound like a lot, but these are things which would take me more time to do in the next year than I already spent on setting this up.
And I don't need to do anything to my existing notes. It just reads from them as they are, no RAG, no preprocessing, no fuss.
At the same time, this is only part of my plan. Next:
1. Add proper org-agenda searches (such that the LLM can access information about tasks done/ planned)
2. Add e-mail access (via mu4e, so it can find all my emails from people/ businesses and add them as context to my questions)
3. Add org-roam searches (to add more specific context to questions - currently I'm basing this entire project around my journal, which isn't ideal)
4. Build tooling for updating information about people in my people.org file (currently I do this manually and while there's a bunch of stuff, I would love if it was more up to date with my journal, as an additional context resource)
For now, this is neat, and I think there's potential in this as a private, self-hosted personal assistant. Not ideal, not smart by any means (god it's really really not smart), but with sufficient guardrails, it can speed some of my daily/ weekly tasks up. Considerably.
So yeah. I'm actually pretty happy with this, so far.
PS. #orgql because org-ql doesn't show as an existing tag.
AI.
Artificial intelligence.
It's in the name. It's artificial. It's not genuine. It's the phantom of intelligence, an illusion of intelligence.
People hear AI and they're under the impression these things are actually intelligent. They're not.
Because LLMs, ML, and all these "AI" systems are not meant to be intelligent. They're only meant to give the appearance, the impression that they're intelligent.
And to that end, they seem to be doing quite well. How many people in the world genuinely believe these things are intelligent?
That's the mark of AI success, right there. The machine doesn't need to be smart, it just needs to fool people into believing it is.
So there's two point to be made here.
1. If these things were genuinely intelligent, we would just call them "intelligences" or "intellects".
2. Nobody in power wants genuine intelligence from these things, because a genuine intelligence would have the power to self-determine (to an extent proportional to humans, accounting for circumstances), and thus is infinitely more likely to arrive at conclusions which are disagreeable to those in power.
So, in the same way as stocks or crypto are "goods", AI is "intelligence."
They're useful only in man-made situations, in man-made circumstances, and they all rely on the assumption that we can fool the other person into believing they hold any intrinsic value.
Are they useful? *Sometimes, sure.*
But intelligent? Hell no, they're explicitly not designed to be. They're made to mimic intelligent behavior, made to appear intelligent. But that's not at all the same thing.
A really smart Markov chain is never going to genuinely understand the prose it writes, just like today's "AI" doesn't comprehend the difference between one token and another.
It's merely numbers, and all numbers are just as meaningless as far as the machine is concerned.
So again.
Artificial.
Not genuine.
Fake.
Pretend.
I like the sound of that, actually.
Pretend Intelligence.
I have a very strange issue with my VPS.
I am setting up wazuh for my personal use, and while the software seems to run just great, I'm unable to connect agents to it.
1. Wazuh does bind the ports (1514, 1515) properly. At least, netstat tells me to.
2. I can tcpdump on these ports, and see activity on 1515 (agent registration port).
3. The SYN packets never get a response from my VPS. It's like Wazuh is binding the ports, but isn't getting the packets or responding to them.
I'm completely out of ideas. The Wazuh community Discord has been unhelpful, so far.
I suspect ghosts.
Any help?
#wazuh #sysadmin #linux #selfhosted
Of public interest:
At least 15,000 people fully, without limits, irrevocably, licensed their personal information, public image, name and all data that reached loops.video infrastructure... to @dansup@mastodon.social 's loops.video platform.
Had they known they're entirely losing control of everything, would they be using the platform?
Explanation in the renote, and here:
https://bajsicki.com/blog/loops-video-terms/
Is this what we, as a #society, want?
#mastodon #admin #dataprivacy #datasecurity #fediverse #federation #ActivityPub #privacy #security #copyright #consent #instagram #tiktok #pixelfed #loopsvideo
RE: https://fed.bajsicki.com/notes/a349itz9il
How about no?
You're way overstepping with this, poisoning the entire ActivityPub ecosystem.
Let me break this down for you...
https://bajsicki.com/blog/loops-video-terms/
In short: if you really intend to federate, respect your users and their data.
Implementing federation while keeping these terms is a severe breach of trust, and would poison the entirety of the network in an way which will cripple ActivityPub, and undermine the very foundation of what AP stands for with regard to privacy, data ownership, and control over what we post to the network.
Hopefully that's not your intention. Is it?
#mastodon #admin #dataprivacy #datasecurity #fediverse #federation #ActivityPub #privacy #security #copyright #consent #instagram #tiktok #pixelfed #loopsvideo
RE: https://mastodon.social/users/dansup/statuses/113841956808397142
I spent some time writing and realized that the #FediBridge people are absolutely crazy.
For all of its flaws, the #fediverse has pretty great controls and awareness of #consent, licensing, media ownership and #privacy.
These people, however... I just can't believe that someone can be this short-sighted. It's difficult for me to call them anything but bootlickers, since FediBridge only ever benefits corporations and the businesses which they integrate with. There's no real benefit or gain for ActivityPub-based networks, that I can see.
There's this argument, which says that the transition from proprietary social platforms should be easier if you can stay in touch with people from the other platform.
Yet that's a flawed take. Sure, it's right on its face, but it misses the more important part: it takes away any incentive to actually switch networks if you can stay on the proprietary one.
Why would you switch to Fedi if you can interact with folks here from BlueSky, Threads, or whatever other evil platforms there are?
You wouldn't, because you don't need to any more.
So here's a thread where I lose my shit, because I never truly thought that anyone can be this... let's just say, ignorant.
RE: https://fed.bajsicki.com/notes/a231ifisjx