iiinova — where your digital self keeps evolving

lewburton.bsky.social @lewburton.bsky.social · 11h

I was in the market for a macropad and current trends couldn't have lined up any better for my taste

lewburton.bsky.social @lewburton.bsky.social · 26/09/2026

"This is called anthropomorphizing, when we ascribe human traits, emotions or behaviors to non-human things, such as animals or inanimate objects." OK AP, my style guide is to not write stupid stuff unless it is intentionally stupid.

000

lewburton.bsky.social @lewburton.bsky.social · 24/09/2026

The best thing service some agent like muse or whatever could provide is this: read and assess social media/news, and tell me "You don't want to be online today, might be safe in 48 hours" That would be very, very helpful

000

lewburton.bsky.social @lewburton.bsky.social · 24/09/2026

I am seeing more people passionately against Pangram and that does not not raise suspicions about motivation for that.

001

lewburton.bsky.social @lewburton.bsky.social · 22/09/2026

I stil don't understand the psychology or psychological profile or whatever of the people who have, say, Claude write for them and pass it on, publish it without considering that everyone else can identify the hallmarks of AI written text—even without using Pangram. Do they not: know, see it, care

100

lewburton.bsky.social @lewburton.bsky.social · 20/09/2026

Muse is designed to give you the lowest hanging fruit available to someone with more agency while training you to not be that person

000

lewburton.bsky.social @lewburton.bsky.social · 17/09/2026

The thing about the p(doom) people is this: they underestimate humans. If there is a misaligned/nefarious ai, or whatever: the human race will eradicate it. Humans have eradicated species before and even viruses. Measles, for example, totally eradicated by vaccination.

011

lewburton.bsky.social @lewburton.bsky.social · 17/09/2026

There have always been v smart people who have believed in the woo, not just any woo, but the pure high-grade woo. I wish I understood that better. At the same time, we now have technology that feels more alchemical than anything else I can recall…best not to have higher-order thoughts about it.

010

lewburton.bsky.social @lewburton.bsky.social · 16/09/2026

I made the mistake of reading the main feed, or discover. Yikes

000

lewburton.bsky.social @lewburton.bsky.social · 15/09/2026

Still thinking about the swarm of duck sized horses. Which is sort of a good way to think about the the difficulty of defense. We are maybe more equipped to deal with one large threat. People say "these things can't even replace SaaS" Ok but recall the damage inflicted by the flea not too long ago

100

lewburton.bsky.social @lewburton.bsky.social · 15/09/2026

An interesting dynamic. AI is coming after law firm fees (stories that banks asked firms to lower billable hours or rate). Also law firms will be the ones coming after the labs. Is it devil take the hindmost? I don't know what alliances will survive.

010

lewburton.bsky.social @lewburton.bsky.social · 15/09/2026

Intelligence is mostly looked down on and not valued highly; I'm no supply/demand expert but I don't see how an abundance of intelligence (a datacenter of geniuses) is going to raise the value of intelligence. $IQQ

000

lewburton.bsky.social @lewburton.bsky.social · 15/09/2026

How much of the AI safety issue is them discovering that a hundred duck sized horses is as bad a as one horse sized duck? How much is the that their success so far is assisted by an absence of legal, compliance, and risk management? How much is that they have no idea what they are building?

110

lewburton.bsky.social @lewburton.bsky.social · 14/09/2026

Also the two strong beliefs: 1. The pause is because the models are stalled 2. The pause is because the unreleased model is too strong

000

lewburton.bsky.social @lewburton.bsky.social · 14/09/2026

So many people giving practically any agent full access to their email and just about everything. These people are nuts. Notable threat vector knower Central Cee: "Before I give you my Insta' password, I'll give you the pin to my Amex"

000

lewburton.bsky.social @lewburton.bsky.social · 14/09/2026

The most interesting proposal going around is that VCs should be financially and criminally responsible for AI companies actions.

000

lewburton.bsky.social @lewburton.bsky.social · 14/09/2026

Fable continues to suck for me. So the idea that Anthropic has cracked scaling or capabilities or whatever, is super attractive as an idea, but elusive as I am stuck in the notches of the jagged frontier. When Claude Fable is not doing knowledge work it can be pleasant to interact with.

100

lewburton.bsky.social @lewburton.bsky.social · 13/09/2026

Last month I was setting up some Hermes agent on a VM and the LLM was like "I accidentally read and API key you need to burn it" and "I don't think it is a good idea for your agent to have access to your email" "Use fine grained tokens" Now Poke Instinct LV Muse all want me to hook up my email.

110

lewburton.bsky.social @lewburton.bsky.social · 13/09/2026

I recall visiting Las Vegas in 2006 and I was in some new residential area and they built the houses but the roads hadn't been finished, stores weren't open yet. This is just a possibly wrong memory But feels like that is the state of the AI companies.

110

lewburton.bsky.social @lewburton.bsky.social · 10/09/2026

I'm being told that Claude will buy Anthropic before the IPO.

000

lewburton.bsky.social @lewburton.bsky.social · 10/09/2026

If this were really true, it does not seem like a safe place to work. We've all seen this movie before. Not at all cool.

000

lewburton.bsky.social @lewburton.bsky.social · 09/09/2026

I wish the AI safety people would focus on Claude making me read base R, instead of threats to humanity which is not at all important in the scheme of things.

000

lewburton.bsky.social @lewburton.bsky.social · 09/09/2026

Everyone is focused on what openAI still keeps for training when you opt out. Simple solution, don't ask anything interesting. I still don't have a clear answer on who has access to my private repos on GitHub. Also has a simple solution but annoying

000

lewburton.bsky.social @lewburton.bsky.social · 09/09/2026

I had a long post about how hedge funds surely have training and policies in place to prevent AI companies from training on them. Fortunately, I think I never posted it because I guess I am wrong. Felt sure, maybe I realized I wasn't. Good for me

000

lewburton.bsky.social @lewburton.bsky.social · 09/09/2026

How resources allocated at openAI and Anthropic? Central planning? Can't imagine there is any P&L that makes sense. Doesn't seem great.

000

lewburton.bsky.social @lewburton.bsky.social · 08/09/2026

"We have de-identified this chat that discussed Mathematica code to predict the optimal age for sauternes in a wine cellar, there are no identifying data here, none"

010

lewburton.bsky.social @lewburton.bsky.social · 08/09/2026

This from openAI is something that I believed possible based on what I have read about models training. What amazed me is influential people who seem to be totally unaware "While unlikely, we cannot rule out that de-identified data derived from their usage of our products helped improve our models"

100

lewburton.bsky.social @lewburton.bsky.social · 08/09/2026

I'm not saying anyone stole anyone else's idea, I am just saying it would be unusual for someone not to.

000

lewburton.bsky.social @lewburton.bsky.social · 04/09/2026

Fable 5.1 left out the $...$ in a md document, took 2 minutes to add them. Only a human would be that dumb and slow, thus AGI is here

010

lewburton.bsky.social @lewburton.bsky.social · 02/09/2026

Fable 5.1 was brutally mogged on finance eval by a cheaper foreign import. Sad.

000

lewburton.bsky.social @lewburton.bsky.social · 01/09/2026

I see someone using real GDP per capita over time and it's like reading one of those armchair epidemiologists posts but macro. Basically, same type of person, different hobbies. Food, health, economics &c.

020

lewburton.bsky.social @lewburton.bsky.social · 01/09/2026

Businesses are going to make sure people hate ai by forcing us to deal with the shittiest models for customer service on their broken vibe coded products (the lamp I just bought). And it isn't the fault of people anthropomorphizing agents, it's stupid people making stupid decisions.

110

lewburton.bsky.social @lewburton.bsky.social · 01/09/2026

I really don't care much about using words like civilization or anthropomorphizing agents; I want (maybe not at all want) to see what happens when you have heterogeneous agents and an economy, one with scarcity. When game theory can we haz. When war.

020

lewburton.bsky.social @lewburton.bsky.social · 30/08/2026

Imagine an endless supply of Yquem and foie gras. I think people of a certain age see the capabilities of computing and robotics—the endless source of knowledge and production as this amazing valuable thing, and for many others, most maybe, zero appreciation and maybe a revulsion.

110

lewburton.bsky.social @lewburton.bsky.social · 26/08/2026

It occurred to me that curing diseases is possibly/probably unaffordable and interventions to prevent disease would be the only solution; demographics, batshittery, the timeline and economics of prevention medicine all point to it not happening in time.

110

lewburton.bsky.social @lewburton.bsky.social · 25/08/2026

AI for writing: it's great for catching problems. But people are behaving badly, using AI merely to save time; the slop cost is borne by the readers, gains captured by the author. One case against data centers is similar: privatized gains and socialized costs. Brevity is kind; slop is an insult.

110

lewburton.bsky.social @lewburton.bsky.social · 25/08/2026

I fear it's macro season again and,…,gross

000

lewburton.bsky.social @lewburton.bsky.social · 25/08/2026

So the high conviction takes this month are that the SALP portfolio was both nothing more than degenerately leveraged beta and entirely composed of insider trading.

000

lewburton.bsky.social @lewburton.bsky.social · 24/08/2026

If scarcity and bottlenecks have legs as a theme across domains: Good for long term strategic actors, not so good for those with an unorganized command each focused on winning individual battles.

000

lewburton.bsky.social @lewburton.bsky.social · 17/08/2026

"Pretty naive take," he says, after misunderstanding what he had read.

000

lewburton.bsky.social @lewburton.bsky.social · 17/08/2026

Lots of smart comments on Jane Street but I am not convinced. I haven't seen this idea yet: Suppose the whole business is 1. naturally short semiconductor prices (because of their planned build); 2. naturally short the market (empirically); 2 makes hedging 1 not so bad in some cases

220

lewburton.bsky.social @lewburton.bsky.social · 16/08/2026

It's not a hedge fund you absolute dimwit

000

lewburton.bsky.social @lewburton.bsky.social · 14/08/2026

Flash 3.7 gave a surprisingly good answer for an everyday non-finance, non-compute sort of question; so I decided to test on a fresh instance, and it is down.

000

lewburton.bsky.social @lewburton.bsky.social · 13/08/2026

It funny to see Nassim get dragged in health, he has always been weak on this topic; I mean more than usual

000

lewburton.bsky.social @lewburton.bsky.social · 07/08/2026

I walked into the middle of this nation's second largest anime/cosplay convention and it isn't even the weirdest thing about my day I am not sure I recognize a single character

000

lewburton.bsky.social @lewburton.bsky.social · 06/08/2026

I found myself performing the fabled GPT Sol loop: Are you sure? Are you sure? Final Answer? It is annoying to me that this works and I need to figure out how to preempt this nonsense. Probably going to get a Fable sub tomorrow. This is going to eat into my foie gras budget.

010

lewburton.bsky.social @lewburton.bsky.social · 05/08/2026

Another Fundsmith add? Gotta be glum over there

000