totototo

wh1le(al1ve){ sh1p_1deas(); r3fact0r(l1fe); }

[totototo @tcs131]

@Hanseatengrau Ja

[totototo @tcs131]

@StephanVZ @Sunnymica Euch? Und wo verteidige ich irgendwelche Milliardäre? hast du Halluzinationen?

[totototo @tcs131]

@StephanVZ @Sunnymica for text in trainingsdaten:
tokens = tokenizer(text)
vorhersage = modell(tokens[:-1])
fehler = loss(vorhersage, tokens[1:])
fehler.backward()
optimizer.step()

[totototo @tcs131]

@StephanVZ @Sunnymica Nein du weisst anscheinend nicht wie ai training funktioniert.„Benutz halt keine KI“ löst nicht das Problem, dass die Trainingsdaten vorher schon ungefragt verwertet wurden. Du argumentierst komplett am eigentlichen Punkt vorbei.

[totototo @tcs131]

@PhilippMattheis dir ist aber schon klar dass du gerade mitmachst bei der aktion?

[totototo @tcs131]

@StephanVZ @Sunnymica es geht darum dass die modelle mit eben diesem individuellem geistigen Eigentum trainiert wurden ...

[totototo @tcs131]

@SprengmEAcc @_kittypunk LLMs halluzinieren, weil sie Text auf Basis von Wahrscheinlichkeiten generieren und Informationen nicht mit einer echten Faktenbasis abgleichen. RAG und Validierung schränken das Problem ein es ist aber nur reduziert nicht gelöst.

[totototo @tcs131]

@SprengmEAcc @_kittypunk Na das hatte ich vergessen :) aber spass beiseite du bist ja mal putzig :)

[totototo @tcs131]

@SprengmEAcc @_kittypunk auch das "denkende" Modell liefert Halluzinationen. selbst schuld wenn man meint man könnte dieses technische problem durch ein abo umgehen ;)

[totototo @tcs131]

Funday

[totototo @tcs131]

@FrankfurtZack Heute ist x wieder besser als Kino 🍿

[totototo @tcs131]

Low level

[Low Level @LowLevelTweets]

there is literally nothing wrong with goto

[totototo @tcs131]

@Kali_im_Exil Juckt jetzt genau wen? 😂 hier werden einem echt strange people in die timeline gespült

[George @0xBaltar]

RDMA for everyone!*

2x Strix Halo 128GB
+
2x Aliexpress USB4 Cables
+
custom Linux kernel module + userspace
=
~236 usable GB VRAM
~40Gbps full duplex bandwidth, single-digit μs latency

versus ethernet:

11x faster finetuning
14 -> 20tok/s tg at batch=1 on MiniMax-M2.7-AWQ

[totototo @tcs131]

@MichaBertram Bei 6 Milliarden euro Kirchensteuer und 3 bis 4 Milliarden personalkosten. Da bleiben schlappe 2 bis 3 Milliarden über. Da Eintritt zu nehmen is ein Armutszeugnis

[totototo @tcs131]

we should move all our repos to codeberg

[vx-underground @vxunderground]

Chat, I don't want to be that guy, but I think Microsoft has really pissed off security researchers and we're approaching the tipping point.

This Eclipse guy has really rocked the boat for Microsoft.

[Andrej Karpathy @karpathy]

LLM Knowledge Bases

Something I'm finding very useful recently: using LLMs to build personal knowledge bases for various topics of research interest. In this way, a large fraction of my recent token throughput is going less into manipulating code, and more into manipulating knowledge (stored as markdown and images). The latest LLMs are quite good at it. So:

Data ingest:
I index source documents (articles, papers, repos, datasets, images, etc.) into a raw/ directory, then I use an LLM to incrementally "compile" a wiki, which is just a collection of .md files in a directory structure. The wiki includes summaries of all the data in raw/, backlinks, and then it categorizes data into concepts, writes articles for them, and links them all. To convert web articles into .md files I like to use the Obsidian Web Clipper extension, and then I also use a hotkey to download all the related images to local so that my LLM can easily reference them.

IDE:
I use Obsidian as the IDE "frontend" where I can view the raw data, the the compiled wiki, and the derived visualizations. Important to note that the LLM writes and maintains all of the data of the wiki, I rarely touch it directly. I've played with a few Obsidian plugins to render and view data in other ways (e.g. Marp for slides).

Q&A:
Where things get interesting is that once your wiki is big enough (e.g. mine on some recent research is ~100 articles and ~400K words), you can ask your LLM agent all kinds of complex questions against the wiki, and it will go off, research the answers, etc. I thought I had to reach for fancy RAG, but the LLM has been pretty good about auto-maintaining index files and brief summaries of all the documents and it reads all the important related data fairly easily at this ~small scale.

Output:
Instead of getting answers in text/terminal, I like to have it render markdown files for me, or slide shows (Marp format), or matplotlib images, all of which I then view again in Obsidian. You can imagine many other visual output formats depending on the query. Often, I end up "filing" the outputs back into the wiki to enhance it for further queries. So my own explorations and queries always "add up" in the knowledge base.

Linting:
I've run some LLM "health checks" over the wiki to e.g. find inconsistent data, impute missing data (with web searchers), find interesting connections for new article candidates, etc., to incrementally clean up the wiki and enhance its overall data integrity. The LLMs are quite good at suggesting further questions to ask and look into.

Extra tools:
I find myself developing additional tools to process the data, e.g. I vibe coded a small and naive search engine over the wiki, which I both use directly (in a web ui), but more often I want to hand it off to an LLM via CLI as a tool for larger queries.

Further explorations:
As the repo grows, the natural desire is to also think about synthetic data generation + finetuning to have your LLM "know" the data in its weights instead of just context windows.

TLDR: raw data from a given number of sources is collected, then compiled by an LLM into a .md wiki, then operated on by various CLIs by the LLM to do Q&A and to incrementally enhance the wiki, and all of it viewable in Obsidian. You rarely ever write or edit the wiki manually, it's the domain of the LLM. I think there is room here for an incredible new product instead of a hacky collection of scripts.

[totototo @tcs131]

@MWirtz30 @Schwaebli1 Spürt ihr noch irgendwas?

[totototo @tcs131]

Nur einige Accounts können ihm antworten dass schon jetzt Immobilien das denkbar schlechteste asset sind. Machen Arbeit Ärger und der Profit liegt unter msci world. Was ein langsam denkender Mensch aber Hauptsache taxtherich :(