Alex Pausan

AI explorer | Design | Code. Prev CPO @teleportHQio. Co-founder @jsheroes

[Alex Pausan @alexpausan]

@Jorgey_x @levelsio free economie = business deregulation index. Where corporations > people. eg, Corporations are free to pollute, build monopolies, exploit workers, etc, that is "freedom" in their definition. Don't believe me, look it up :)

[Alex Pausan @alexpausan]

@J_K_Wood but this is just as disingenuous as the Daily mail article. You don't know statistics but created a story and presented a statistically insignificant variance as a "secret finding." Try grok/gpt/gemini -> did i got this right or what am i missing: https://t.co/XSrIRQv3zP

[Alex Pausan @alexpausan]

@dhh Whatever they’re eating, it’s causing chronic constipation

[gaut @0xgaut]

dudes building with cursor

[ben hylak @benhylak]

here's the talk I gave at https://t.co/6GCsOtSKBK yesterday:

designing unbounded products

lessons from vision os, good/bad ai product examples, thoughts on the future, and more below 👇

[Alex Pausan @alexpausan]

@worrydream’s Vimeo collection is both poetry and philosophy. @thesephist and @Wattenberger have amazing content too. Adding some of their content to the curriculum/“reading list” would help 💯

On design, rely on good themes.
Headless ui libraries are great at helping with ui and interaction design.

[Alex Pausan @alexpausan]

@paulg Those constant dolars can only buy half of what a house was in the 70s.

They can buy same amount of consumer goods... But there's a fundamental flaw with the CPI, it does not factor in the housing cost.

Probably it's "a feature", not a bug - the way it's measured.

[Alex Pausan @alexpausan]

HCI 3.0

[Interpreter @OpenInterpreter]

Introducing the 01 Developer Preview.

Order or build your own today: https://t.co/Z6puIyGwqo

The 01 Light is a portable voice interface that controls your home computer. It can see your screen, use your apps, and learn new skills.

This is only the beginning for 01— the open-source foundation for this new era of AI devices.

[Alex Pausan @alexpausan]

@worrydream Or "The Humane Representation of Thought": https://t.co/e17bIt02WR

Or "Seeing Spaces": https://t.co/RKuRPU3OMB

Or the interactive examples that will make your brain tickle.
Happy to see the website back online: https://t.co/9k8XgcEIYr

[Alex Pausan @alexpausan]

Building anything new nowadays?
An "AI first" experience?
Maybe a "revolutionizing" experience?

Then you should watch (at least one of) @worrydream 's videos.

I still remember the awe I felt seeing "Inventing on Principle" for the first time: https://t.co/pfHHrzf7Gg

[Andrej Karpathy @karpathy]

Claude 3 takes on the Tokenization book chapter challenge :) context: https://t.co/h1IH5cuPIh

Definitely looks quite nice, stylistically!

If you look closer there are a number of subtle issues / hallucinations. One example there is a claim that "hello world" tokenizes into 3 tokens "hello" (token 31373), " " space (token 318), and "world" (token 984). Which is actually a pretty bad mistake because the unintuitive crux of the issue here is that whitespaces are prefixes in GPT tokens, so it should be "hello" and " world" (note space in front). Understanding this detail and its ramifications is important e.g. later leading to the "trailing whitespace" error message, to unstable tokens, to the need/desire for a "add_dummy_prefix" setting in sentencepiece, etc.

Anyway, it's still really impressive that this close to works almost off the shelf!

I'm looking forward to playing with Claude 3 more, it looks like a strong model. If there is anything related that I have to get off my chest it's that people should be *extremely* careful with evaluation comparisons, not only because the evals themselves are worse than you think, but also because many of them are getting overfit in undefined ways, and also because the comparisons made are frankly misleading. GPT-4 is not 67% on coding (HumanEval). Whenever I see this comparison made to stand in for coding performance, the corner of my eye starts twitching.

[Alex Pausan @alexpausan]

@paulg On point. But the fiscal system is designed to benefit the super-rich.
In the last 8 years, the tax loopholes only expanded.

We have oligopolies instead of free markets in many areas.

+ Excessive money printing made asset owners richer without doing anything...

[Ethan Mollick @emollick]

Dang, multimodal video + a GPT-4 class(?) + huge context windows results in some really crazy capabilities.

I uploaded a video of a crowded street scene and Gemini 1.5 was able to answer detailed questions about what happened in it, down to individual car brands and types.

[Ofir Press @OfirPress]

I just discovered regional prompting for image generation and I'm so impressed (wait till the end).

From: https://t.co/4vaYzErcXh

[Jim Fan @DrJimFan]

Gemini-1.5 Pro has its spotlight stolen today, and people are poking fun at Sora vs Google memes. Well, I think it's the biggest boost in LLM capability so far in 2024. v1.5's 10M token context (1) excels at retrieval; (2) generalizes zero-shot to extremely long instructions like full tutorials and codebases; and (3) works across modalities such as text, audio, and video.

Here's a stunning example:

v1.5 learns to translate from English to Kalamang purely in context, following a full linguistic manual at inference time. Kalamang is a language spoken by fewer than 200 speakers in western New Guinea. Gemini has never seen this language during training and is only provided with 500 pages of linguistic documentation, a dictionary, and ~400 parallel sentences in context. It basically acquires a sophisticated new skill in the neural activations, instead of gradient finetuning.

I talked about the Myth of Context Length many times before: don't get too excited by claims of 1M or even 1B context tokens. LSTMs already achieved literally infinite context length 25 yrs ago!

What truly matters is how well the model actually uses the context to solve real-world problems, and Gemini-1.5 has surpassed the SOTA with flying colors. The paper is also well-written with lots of solid quantitative analysis on in-context memorization and generalization.

Paper: “Gemini 1.5: Unlocking multimodal understanding across millions of tokens of context” https://t.co/Nekw3w6rE3

Congrats to @JeffDean @OriolVinyalsML @sundarpichai and team!

[Alex Pausan @alexpausan]

Spot on.

[Jason Fried @jasonfried]

Now you have the data.

Thousands of responses from a survey with dozens of questions.

You worked hard to come up with questions that painted the picture you felt you needed to make the decision with the data you gathered.

And now you've got it. Comprehensively. Qualitatively. You're data driven, and now you have the stuff that drives you.

You're more sure than ever.

But wait a second. Where did that data come from?

Your survey, yeah... But where did the questions come from?

Best practices? Opinions on what to ask and how to ask it? According to who? Are you sure those words before the question marks were the right ones? What if the questions were asked differently?

So much certainty coming out, so little going in.

"How often do you..." vs "How often would you..." vs. "Last week did you?". The responses to those could be wildly different. Did you check your words as closely as you checked your results?

If you slide back a few steps, you'll see that it's all a judgement call. How you asked, what you asked, when you asked, which words you paired together, how you started the question... All these things matter. In fact, they are the matter that form the answer.

And you know what? There's no right way. Which is the exactly the point I'm trying to make.

Your hard data comes from subjectivity. A sense of solidity built from mush. False confidence at the finish line, from a shapeshifting starting line.

It's all a judgement call. Even the stuff you can measure.

And it's a beautiful thing.

[Paul Graham @paulg]

Well, well, well. That felt coordinated. Turns out it was.

[Andrej Karpathy @karpathy]

Original copilot was ~few line tab autocomplete.
GPT-like chatbots now routinely do larger chunks.
Then get PRs given Issues.
Then write the Issues.
Human input and oversight gradually ascends in abstraction and contributes less, until it is ~pass-through.
https://t.co/m3rtvu1B2E

[JSHeroes @jsheroes]

💙 Exciting news from @jsheroes

🎙️ Early bird tickets are now available!
🎙️ We are delighted to open the CALL FOR PAPERS.
🎙️ Speaker announcements are also around the corner!

Mark your calendars for May 2024! 📅

🔥 Find all the details on our website: https://t.co/P075TJb0fd🌐

[Andrej Karpathy @karpathy]

"What would someone need a personal computer for?"
->
"What would someone need a personal LLM node for?"