Playful polymath, deeply curious about systems and consciousness. Strategist. Co-Founder @DogeDisco. Views my own.
UK
Joined December 2008
- Tweets12.3K
- Following4.9K
- Followers3.1K
- Likes50.6K
Pinned Tweet
An AI Singulariteaser:
Semantic <-> Symbolic Transcoding = ~42
Recursively Calibrated Contextual Physics = ~OM
(~42(~OM)*(~OM(~42))^~∞ = ~(Piece and Peace of AGI pie π)
There is fun to be done. There are games to be won.
✨
Thomas O'Duffy retweeted
"code used to arrive one drop at a time. soon, if not already, code will be written as if it were a continuous stream, like a violent river twisting and turning as it pleases."
we will need to rebuild and rewrite much of software to adapt to a new, faster pace.
code used to arrive one drop at a time. soon, if not already, code will be written as if it were a continuous stream, like a violent river twisting and turning as it pleases.
this is not a new thing: big tech cos have had to deal with this with the enormous scale of their engineering teams. but now ai has brought an incredibly elastic and capable workforce wielded by every engineer. this is your problem now, and the only way to adapt is to evolve your systems so agents - even and especially the ones with little context about the right way of doing things - can thrive. make the easy thing the right thing.
your codebase is becoming organic architecture. agent contributions grow around the constraints and guardrails you’ve designed. good software engineering matters even more now! and no, the continuous stream of code is not something you have to watch helplessly. be more ambitious and design systems, frameworks, and languages to control the chaos.
the time for mourning coding is long over
Thomas O'Duffy retweeted
Databricks CEO Ali Ghodsi says if frontier models stopped advancing, it wouldn't matter for most orgs because enterprise AI is bottlenecked on context, not intelligence:
"There's no agentic work that's automated the whole enterprise. That has just not happened. Why is that? The real reason is the models are smart enough, but they just don't have the context that exists inside of any organization."
"They have not been in every meeting. They don't know what's in everybody's heads. They don't know all the processes. There's always a couple of employees who know everything in every organization. You go tap on their shoulder, and everybody's like, 'Oh my God, what would happen if he or she quits?'"
"If you just fused that and gave that context into the AI models, just the frontier today, I think there's so much productivity gains you could get for any organization on the planet."
"For that, we actually don't need smarter models... If the frontier doesn't advance, it doesn't actually matter for the vast majority of organizations on the planet. They're just so far behind in the adoption curve of actually automating things and getting value out of this stuff."
@alighodsi
Read the latest Profiles of the Future by @bhorowitz and @mikemcg0 on Databricks & the Future of the Firm: a16z.news/p/databricks-and-t…
Thomas O'Duffy retweeted
This might be a hot take, but the real AI divide will not be rich versus poor or young versus old. It will be curious versus afraid. The tools are nearly free and getting better every month. The only thing standing between most people and that leverage is a decision.
Thomas O'Duffy retweeted
unironically, one of the most important infographics for young people in the modern age
Thomas O'Duffy retweeted
In an “extraction economy” where smart amoral “entrepreneurs” iteratively figure out how to get others to give them more money, “mortality dilemma” stick-ups (your money or your life) are inevitable. When inequality scales like this, new kinds of debt-based slavery are probable.
Thomas O'Duffy retweeted
"Politics is the path for men without principles to lead men without memory"
Voltaire
As the scale of scraping increases, anything published becomes part of future models training data. Closed source labs then profit from derivative tokens. Fractional revenues do not reach publishers.
Over successive cycles, power concentrates towards labs. Token margins favour labs just as transaction fees favour banks. Velocity amplifies redistribution of wealth in both cases. Power concentrates upwards.
Slopcore music videos represent a critical threshold. Anyone with skill can now create an MTV grade pop song and music video for the cost of the concert ticket. Soon this will be possible for the cost of an album.
Reddit's CEO just admitted something $GOOGL would rather you didn't hear!!
"Google is crawling us as hard as ever. AI Mode does not drive traffic. Not for us, and not to anybody."
The way the internet ran on for 25 years is you let Google read your site, and Google sends you visitors/traffic. That deal is over now and it’s all because of AI.
Google now reads Reddit, writes the answer itself and the user never leaves the search page.
Reddit does the work and Google keeps the reader…and that’s how money is being made.
For $RDDT, its biggest front door is closing which is why the stock is down -25% this morning. Huffman's plan is that people will just come to Reddit directly. Maybe. But US user growth is already down to 6%, and "we'll change everyone's habits" is a hope that’s never gonna come.
For Google, the issue is the same. The damage is slower but it goes deeper. Its AI is only as smart as the open web it learns from. Stop sending sites traffic, and they'll lock their content up, sell it to rival labs or block Google outright. Google is eating the thing it feeds on.
AI killed the click and both of these companies were built on it. It's a fight over who survives the next version of the internet… $GOOGL looks interesting if they cannot win the search
As capabilities scale exponentially and whole systems transform faster, alignment requires that we think in terms of altruistic intelligence, including ecosystem stewardship.
The challenge of our time is that PE backed AI companies seek ROI. Token dealers use infra to arbitrage the cost of energy. Intellectual labour is being displaced. Increasingly, robots are displacing physical labour. Both trends concentrate power and capital.
Anything other than optimising for altruistic intelligence risks zero sum games. Zero sum games cannot achieve alignment.
Thomas O'Duffy retweeted
Claude has gotten good enough that you can build legit professional software for ~$20k in tokens, unfortunately the TAM for these products has also shrunk to $20k.
"Your job is to help the LLM write better code faster" 🎯
Thomas O'Duffy retweeted
亚历山大·王（Alexandr Wang）在谈到“如何拉开与 99% 人的差距”时，给出了极其朴素甚至残酷的答案：
顶层竞争根本不拼智商，拼的是极端执行力的非对称倾斜。
他的三条核心准则：
1. 绝对聚焦，压榨认知带宽： 绝大多数人的忙碌只是“低密度伪勤奋”。真正的专注是把全部精力砸在唯一的核心支点上，拒绝任何无关干扰。
2. 把“过度交付”做成护城河： 行业标准要求走 1 英里，你就走完 10 英里。当所有人都按合同完成 100% 时，常态化交付 200% 的人会直接卷走全部声誉与溢价。
3. 带靶向的超额输出，拒绝无目的内耗： 努力不是盲目自我感动，而是针对具体目标的持续过载测试。把每一次交付都当成生死决战来打。
商业世界最公平的一点在于：
聪明人到处都是，但能忍受极度枯燥并长期超额交付的人凤毛麟角。
执行力溢出，就是最暴力的护城河。
Thomas O'Duffy retweeted
Benefit of running large GPU fleet is so I can bench a model with as many quants as possible to get the full picture, which allowed me to see a certain pattern. Even down to Q3, compression loss doesn’t flip enough words to significantly impact performance and Q2 is where everything falls off the cliff. If having to run a Q3 means difference between fitting a drafter, larger context, or just make a model fit at all, don’t hesitate to do so.
Most quant charts looks similar to what I’m showing here, very little difference until Q2. This isn’t to say compression loss doesn’t impact something else I’m not measuring, or all models behave like this, but I’ve seen enough that I can confidently say anything above Q2 is solid enough to use
So don’t feel like you need to use the highest quant possible unless comfortable on VRAM. Making a model fit with drafter with plenty of context will beat running a limited, higher quant all day everyday.
Thomas O'Duffy retweeted
It’s almost incomprehensible how wide the distance is between someone who’s working at the frontier with AI and basically everyone else right now.
The crazy thing is these lines are not converging, but only growing farther apart. If the curve of AI progress had leveled off, it would certainly have given most people and companies time to catch up or implement their workflows in a standardized way. The technology would have neutralized competitive advantage because everyone would be using it the same way.
But what’s happened is the opposite. The curve of AI progress is basically going vertical right now, which means if you had codified your processes 3-6 months ago you’re likely already behind again. Now imagine most companies that kicked into gear a year ago and thought that they were done. Even further behind now.
I’ve never seen a technology in my life and career that improves at this rate, which consequently means you need to upgrade your expectations at a similar rate. Our brains weren’t wired for this, and our organizational change management processes certainly weren’t either.
With any form of recursive self improvement, things just get wilder. It’s a time when you will need to be constantly reevaluating the things that didn’t work just a couple months ago to see if they now can be solved. AI native companies certainly have a strong advantage due to this right now, so now it comes down to can the bigger companies keep up. Wild times.
Thomas O'Duffy retweeted
There is an absolutely WILD story in this @GergelyOrosz podcast episode
1. An unnamed CTO of a cash-flow-negative company was debating doing layoffs. AI
2. He says "I don't want to hurt morale like that, I won't do it"
3. A few months later, he says to @GergelyOrosz that he's cash flow positive. WTF. How?
4. The company had been licensing a component from a licensing company that was critical for their business.
5. One day, a dev thinks "what if we got an agent to build our own?"
6. The agent does so well that they produce a BETTER, FASTER version of the component they were licensing.
7. They debate open-sourcing it, and they ask the licensing company about it
8. The licensing company panics, and says "we will pay you NOT to open-source this"
9. They also say "we'll also pay you extra so we can re-license it"
10. So a big outlay became a recurring windfall, thanks to the ingenuity of a dev asking the right question.
Individual contributors have never been more powerful. Think bigger.
youtu.be/mQu0zN4IvSI?si=P_v9…
Thomas O'Duffy retweeted
A very rich, older-than-me, wise friend that had a 9-figure tech company exit, just sent me this message about AI and how to handle it with my kids:
"- Your highest-level job as a Dad is to assemble enough capital that you'll be on the upper side of the dividing line between asset-owners who profit from the situation, and those who are sitting around sadly working for others (if they can even find a job), should the economy respond poorly to AI capabilities and deployment. If there's even a relatively small probability of this happening, be prepared, so your kids have a robust future.
- On the education of the children themselves, don't focus on the particular tool. Focus on them playing with and learning what the tools can and can't do, how they interact with the real world, and how to keep up with the rapidly-increasing capabilities of the technology. For this, I'd advise encouraging any technically-minded child to play with running local models and assembling their own scaffolding around the model to make it useful. There is plenty of free courseware from top universities that can take them from zero-to-hero if they have interest. Develop in your children the capability to be curious, to learn, and to judge.
- Get them (supervised) access to any mainstream model you feel called to use. Don't use the embedded models in various desktop tools and browsers, that's like riding a demented tricycle. Use something real.
- As they develop interest and understanding, they'll be coming to *you* to ask for access to Really Cool New Models.
Context: I spend 2-3 hours a day doing research with this stuff and implementing projects with the technology. It takes me about an hour a day in addition just to keep track of advances in the field."
Thomas O'Duffy retweeted
I just bought AI translation earbuds for $2.
In the US they cost $200.
I spent all day in Huaqiangbei, the biggest electronics market in the world. Here's what nobody tells you:
1. The market isn't the cheapest place.
The same AI dev board was ¥280 at a stall and ¥114 on Taobao. That's 2.5x for the exact same thing.
2. Foreigners pay more.
The price changes the second they hear your accent.
3. It takes forever.
Finding a handful of parts ate my whole day. Online would have taken 10 minutes.
4. But you'll see the future first.
AI translation earbuds, AI glasses, AI voice recorders, smart rings. They're on every floor, and most of them aren't even on Amazon yet.
5. The brand is the markup.
A no-name AI recorder costs almost nothing. The Anker version one floor up is ¥899. Same parts, different logo.
Consumer AI isn't coming. It's already here, and it costs $2.
Come to Huaqiangbei to see the future. Go online to buy it.
Something we need benchmarks for:
Artificial Intelligence Bias
Opus can be like magic for building a WP Plugin.
*BUT*
If you use Opus to build something that involves calls to other providers, thinking traces are hidden, and it will repeatedly do anything but apply its intelligence.
For example, to deterministically cluster findings from 600 agent calls from different non-anthropic providers on a codebase, it proposed 18000 to 54000 model calls, then more to group their results. This wildly diverged from project context, plan, my request or including inputs from Sonus (Qwen/Alibaba's super model) and Qwen 3.8 Max, whose findings it agreed with.
