iiinova — where your digital self keeps evolving

Max Caldwell @growth.wtf · 23/01/2025

0/3 + "you need help"

Max Caldwell @growth.wtf · 01/01/2025

It's crazy to me just how consistently wrong the google search llm is. Is this like, a quantized gemini 1 that can't even see the search results it is summarizing? That's the only explanation for why doesn't it seem to get even context from featured snippets that they should have cached already.

000

Max Caldwell @growth.wtf · 06/12/2024

Sure, here's an example: huggingface.co/datasets/Ple... I am no expert but it seems legit?

The composition of the dataset adheres to the criteria for public domain works in the EU and, consequently, all Berne-countries for EU authors: any publication whose author is dead for more than 70 years. Additionally, the initial consolidation of public domain status for cultural heritage operates in the EU under the 2019 Copyright Directive (art. 14).

As of June 2024, to limit rights verification, we have retained exclusively titles published prior to 1884.

The corpus will be expanded at a later stage to encompass late 19th century and early 20th century publications, after checking for public domain validity.

010

Max Caldwell @growth.wtf · 05/12/2024

Something I've always been curious about with social networks is how much different actions lead to an audience. At scale, it would be interesting to learn how large accounts got large, but I'm also curious what gets people to follow. Is seeing a good post more influential? Or follow-backs? Well...

Alt text: Two posterior distribution plots displayed side by side. The left plot shows "Posterior ATE of 1 Follow on Following" with a mean of 0.4021 and 94% HDI ranging from -10.25 to 10.23, displayed as a bell curve centered near zero. The right plot shows "Posterior ATE of 1 Post on Following" with a mean of 2.759 and 94% HDI ranging from -6.404 to 12, also displayed as a bell curve. Both plots use a blue line to show the distribution and include x-axes ranging from approximately -30 to 40 and -20 to 30 respectively.

100

Max Caldwell @growth.wtf · 05/12/2024

From 2022 - tsunami danger zones in the bay area per the mercury news: www.mercurynews.com/2022/10/12/m...

010

Max Caldwell @growth.wtf · 05/12/2024

@alpindale.bsky.social Getting there, not quite getting the model injection code right yet. I have to start my actual work now but when or if finish this I'll open source.

000

Max Caldwell @growth.wtf · 05/12/2024

OK, got a little DND game running with github.com/ag2ai/ag2 -- my review is that it's a neat set of concepts but very difficult to use and quite buggy. and I was hard-pressed to think of an actual use case that would fit into this.

010

Max Caldwell @growth.wtf · 05/12/2024

Hey @alpindale.bsky.social , I saw you're working on the ruliad project? Can you help explain how to get the model to follow scripted directions? I'm trying to write an ollama template to try to get it to use tools, or follow scripted tasks of any order.

110

Max Caldwell @growth.wtf · 05/12/2024

deepthoughts-8b-tools is confused, but it's got the spirit?

000

Max Caldwell @growth.wtf · 05/12/2024

Unfortunately it's just not trained for this. Even if you get it working with Ollama tool calling, the model is providing a total hallucination about half the time. and is unable to provide responses in standard tool calling format.

000

Max Caldwell @growth.wtf · 05/12/2024

Tell me you're overthinking it without telling me you're overthinking it. Full reasoning chain in the alt text.

$Hello The user has provided a simple greeting 'Hello'. This is not a complex problem, but rather a basic interaction. I need to understand the context and determine the appropriate response." }, { "step": 2, "type": "data_gathering", "thought": "I have all the necessary data in the user's input: a single word 'Hello'. There's no additional information or context provided." }, { "step": 3, "type": "analysis", "thought": "The greeting 'Hello' is a common English phrase used to greet someone. It's a polite and casual way to initiate a conversation." }, { "step": 4, "type": "hypothesis_formation", "thought": "Given that this is likely the start of a conversation, the most appropriate response would be to acknowledge the greeting and reciprocate with another friendly greeting." }, { "step": 5, "type": "solution_proposal", "thought": "I should respond with 'Hi!' as it's a simple, friendly greeting that matches the tone of 'Hello'." }, { "step": 6, "type": "alternative_consideration", "thought": "Other possible responses could include 'Hey there!', 'Welcome!', or just 'Hi' without an exclamation mark. However, 'Hi!' remains the most straightforward and widely applicable option." }, { "step": 7, "type": "evaluation", "thought": "The proposed solution of 'Hi!' is appropriate because it's brief, friendly, and doesn't assume any additional context that isn't provided in the original message." }, { "step": 8, "type": "decision_making", "thought": "Based on the analysis and evaluation, I will proceed with the response 'Hi!' as it best fits the given scenario." }, { "step": 9, "type": "verification", "thought": "Verifying the reasoning chain: The problem was correctly identified, the context analyzed appropriately, and the solution proposed aligns well with the expected behavior for a simple greeting exchange." },$

020

Max Caldwell @growth.wtf · 05/12/2024

20% worse than o1 and claude-latest at reasoning evals, but better than Qwen-72b, which is interesting?

100

Max Caldwell @growth.wtf · 04/12/2024

Update, tailwind4 totally kicks ass. This is all you need to create new utility classes usable in code... normal css and variables!

030

Max Caldwell @growth.wtf · 03/12/2024

My cat has gotten in the habit of coming to yell at me until I let him sit on my chest around this time. Funny wat to be forced to take a work break. Not allowed to say no to the cat.

Close up of an orange cat looking into camera

110

Max Caldwell @growth.wtf · 28/11/2024

Do you find 300 characters limiting? Would you like to express yourself in way more than a single post can contain? Use the alt text! It can contain almost 6x as many characters before you hit the size limits. Click the alt text to see Canterbury Tales. the first 2,000 characters

Whan that Aprille with his shoures soote,
The droghte of March hath perced to the roote,
And bathed every veyne in swich licóur
Of which vertú engendred is the flour;
Whan Zephirus eek with his swete breeth
Inspired hath in every holt and heeth
The tendre croppes, and the yonge sonne
Hath in the Ram his halfe cours y-ronne,
And smale foweles maken melodye,
That slepen al the nyght with open ye,
So priketh hem Natúre in hir corages,
Thanne longen folk to goon on pilgrimages,
And palmeres for to seken straunge strondes,
To ferne halwes, kowthe in sondry londes;
And specially, from every shires ende
Of Engelond, to Caunterbury they wende,
The hooly blisful martir for to seke,
That hem hath holpen whan that they were seeke.

Bifil that in that seson on a day,
In Southwerk at the Tabard as I lay,
Redy to wenden on my pilgrymage
To Caunterbury with ful devout corage,
At nyght were come into that hostelrye
Wel nyne and twenty in a compaignye
Of sondry folk, by áventure y-falle
In felaweshipe, and pilgrimes were they alle,
That toward Caunterbury wolden ryde.
The chambres and the stables weren wyde,
And wel we weren esed atte beste.
And shortly, whan the sonne was to reste,
So hadde I spoken with hem everychon,
That I was of hir felaweshipe anon,
And made forward erly for to ryse,
To take oure wey, ther as I yow devyse.

But nathelees, whil I have tyme and space,
Er that I ferther in this tale pace,
Me thynketh it acordaunt to resoun
To telle yow al the condicioun
Of ech of hem, so as it semed me,
And whiche they weren and of what degree,
And eek in what array that they were inne;
And at a Knyght than wol I first bigynne.

A Knyght ther was, and that a worthy man,
That fro the tyme that he first bigan
To riden out, he loved chivalrie,
Trouthe and honóur, fredom and curteisie.
Ful worthy was he in his lordes werre,
And thereto hadde he riden, no man ferre,
As wel in cristendom as in hethenesse,
And evere honóured for his worthynesse.
At Alisaundre he was whan it was wonne;
F…

010

Max Caldwell @growth.wtf · 27/11/2024

He blocked me after this :-(

110

Max Caldwell @growth.wtf · 23/11/2024

Explaining the watershed algorithm

000

Max Caldwell @growth.wtf · 20/11/2024

Stack blitz isn't mobile optimize and clicking on get started just leads to this 🤷

100

Max Caldwell @growth.wtf · 19/11/2024

seriously though if you do this I will fight you.

Image with header saying "New Keyboard shortcuts" describing how command-K is used for opening actions, and create link is something else.

111

Max Caldwell @growth.wtf · 17/11/2024

For large language models, seeing that OpenAI codex in 2021 could generate code got me interested in enough to sign up for their API beta. Pre-LLMs, the fast.ai course from Jeremy Howard is where I learned how to actually use neural nets, really great course still.

010

Max Caldwell @growth.wtf · 14/11/2024

The AI safety people WILL NOT let me eat moldy smoked salmon.

Image of user asking Claude if it is safe to eat. Moldy smoked salmon.

110