iiinova — where your digital self keeps evolving

Daniel Eth (yes, Eth is my actual last name) @daniel-eth.bsky.social · 18/03/2026

If anyone’s wondering what Marc Andreessen is up to, he recently posted this on Twitter:

Daniel Eth (yes, Eth is my actual last name) @daniel-eth.bsky.social · 18/03/2026

In a large upset, LTF (the OpenAI-Andreessen super PAC) takes a major loss in IL-02, where they backed Jesse Jackson Jr. Notably Jackson is famously corrupt, and I wonder if LTF’s toxic AI money fed into existing negative sentiments towards him.

140

Daniel Eth (yes, Eth is my actual last name) @daniel-eth.bsky.social · 11/01/2026

Third, the curve for Claude Opus 4.5 is “flatter” than previous models (it does relatively better at longer tasks compared to shorter). And the longest tasks it does are ones where it’s getting ~50%, b/c METR doesn’t have enough tasks that are long enough in their dataset…

150

Daniel Eth (yes, Eth is my actual last name) @daniel-eth.bsky.social · 11/01/2026

Second, on a log plot, note this is hardly above trend. Sure, it *could* represent a new trend, but it seems like every time there’s a model release that overperforms people think timelines get super short, & every time a model underperforms they think timelines get super long…

150

Daniel Eth (yes, Eth is my actual last name) @daniel-eth.bsky.social · 11/01/2026

A few thoughts on Claude Opus 4.5: First off, in absolute terms, this is a pretty big step up. Anthropic is showing they have juice, and things are going faster than previously expected. At the very least, this should dispel all recent talk about how AI was entering a slowdown

2151

Daniel Eth (yes, Eth is my actual last name) @daniel-eth.bsky.social · 05/01/2026

Notably, the public has also shifted away from Republicans on the issue (up on the graph), coinciding with many Republicans pushing an anti-regulatory attitude towards AI. Voter now trust Dems & Rs about equally on the issue, indicating voters are up for grabs by either party

140

Daniel Eth (yes, Eth is my actual last name) @daniel-eth.bsky.social · 05/01/2026

There’s still a long ways to go before AI is a top voter concern like health care or cost of living, but I’d expect this trend to continue as AI becomes more powerful. Politicians who side with wealthy tech donors over voter preferences may wind up regretting that decision

140

Daniel Eth (yes, Eth is my actual last name) @daniel-eth.bsky.social · 05/01/2026

Graph showing AI becoming higher salience to voters (more to the right on the graph). According to this data, AI is now higher salience than climate change, and approaching the salience of gas prices

173

Daniel Eth (yes, Eth is my actual last name) @daniel-eth.bsky.social · 24/11/2025

For those who aren’t following the details, here are the relevant connections:

030

Daniel Eth (yes, Eth is my actual last name) @daniel-eth.bsky.social · 24/11/2025

I think more OpenAI employees should be aware of the very bad-faith political activities that OpenAI is supporting through Greg Brockman’s funding of the Andreessen-OpenAI super PAC cluster (Twitter’s location verification has a known bug, but Leamer doesn’t care about the truth.)

170

Daniel Eth (yes, Eth is my actual last name) @daniel-eth.bsky.social · 21/11/2025

Viewing the graph on a linear scale demonstrate that claims of AI "hitting a wall" are clearly off. People *keep making* these claims, but while not every model release lives up to hype, no, AI has not hit a wall yet, and there's no indication it's about to, either

151

Daniel Eth (yes, Eth is my actual last name) @daniel-eth.bsky.social · 21/11/2025

Things are looking smoothly exponential for AI over the past several years, and I continue to think this is the best default assumption (until the AI R&D automation feedback loop eventually speeds everything up)

3162

Daniel Eth (yes, Eth is my actual last name) @daniel-eth.bsky.social · 19/11/2025

The guy who mocks the pope and funds gambling and porn apps thinks a ban on state AI laws is great! Seems like basically every normal person hates this idea - hopefully politicians recognize it for what it is and don’t side with Andreessen

180

Daniel Eth (yes, Eth is my actual last name) @daniel-eth.bsky.social · 15/11/2025

If you jog in these sneakers it’s called a training run

0200

Daniel Eth (yes, Eth is my actual last name) @daniel-eth.bsky.social · 09/11/2025

Andreessen really doubling down on mocking Catholics

050

Daniel Eth (yes, Eth is my actual last name) @daniel-eth.bsky.social · 09/11/2025

Andreessen is so dogmatically against working on decreasing risks from AI that he’s now mocking the pope for saying tech innovation “carries an ethical and spiritual weight” and that AI builders should “cultivate moral discernment as a fundamental part of their work”

0171

Daniel Eth (yes, Eth is my actual last name) @daniel-eth.bsky.social · 09/08/2025

GPT-5 didn’t live up to OpenAI’s hype, but it is *exactly* in line with extrapolations from prior AI advancements. Go ahead and discount future statements from OpenAI/Altman, but you should still expect the fast AI progress that we’ve been seeing to continue

1274

Daniel Eth (yes, Eth is my actual last name) @daniel-eth.bsky.social · 08/08/2025

Incremental improvement along the same exponential trend (compare GPT-5 to o3):

120

Daniel Eth (yes, Eth is my actual last name) @daniel-eth.bsky.social · 03/08/2025

Looks like from a mental health perspective, you want to make sure to do at least ~2hr/wk of light exercise (eg jogging) or ~1hr/wk of vigorous exercise, and there’s not much benefit to going beyond that.

3131

Daniel Eth (yes, Eth is my actual last name) @daniel-eth.bsky.social · 25/07/2025

Woah Zuckerberg has a $300M estate in Hawaii?! That’s like 1/3 the cost of an AI researcher!

050

Daniel Eth (yes, Eth is my actual last name) @daniel-eth.bsky.social · 26/05/2025

It’s been an entire 5 weeks - I think we need an update to this chart

070

Daniel Eth (yes, Eth is my actual last name) @daniel-eth.bsky.social · 26/05/2025

Surprised and pleased to see this - Dario (Anthropic CEO) hints that he’s against the proposed 10 year state-level AI regulation ban:

2153

Daniel Eth (yes, Eth is my actual last name) @daniel-eth.bsky.social · 25/05/2025

On the first point - Epoch finds that in language models, pretraining algorithmic progress has been around half as impactful as compute scale up. Naively, if compute scale up stopped, progress would slow down by 3x. This is a decent amount, but not enough to say “2030 or bust”

181

Daniel Eth (yes, Eth is my actual last name) @daniel-eth.bsky.social · 25/05/2025

Guy on the right is sad he can no longer use a computer, wishes he had spent more time working on his b2b SaaS startup when he was younger

3273

Daniel Eth (yes, Eth is my actual last name) @daniel-eth.bsky.social · 23/05/2025

Every now and then I think of this quote from AI risk skeptic Yann LeCun

2111

Daniel Eth (yes, Eth is my actual last name) @daniel-eth.bsky.social · 21/05/2025

I feel like this is an unreasonable expectation for the product

1191

Daniel Eth (yes, Eth is my actual last name) @daniel-eth.bsky.social · 19/04/2025

One thing a lot of people missed from AI 2027: Yes, the pause was necessary for surviving. But it only worked because it was ~perfectly timed - after ~AGI but before ASI. If it had instead been earlier, America would have ceded its lead for nothing and we would have all died.

2100

Daniel Eth (yes, Eth is my actual last name) @daniel-eth.bsky.social · 27/03/2025

Having looked into evidence about relevant dynamics, as well as the main arguments against a software intelligence explosion, we think this one could go either way.

120

Daniel Eth (yes, Eth is my actual last name) @daniel-eth.bsky.social · 27/03/2025

COUNTERARGUMENT 2: Progress will be slowed by the time required to train the next generation of AI systems. If each major advance requires a multi-month training process, then you can’t fit arbitrary advances within several months.

110

Daniel Eth (yes, Eth is my actual last name) @daniel-eth.bsky.social · 27/03/2025

COUNTERARGUMENT: Maybe fast software progress depends on rapidly growing hardware for running AI experiments? In this case, hardware could still be a bottleneck, preventing a software intelligence explosion.

110

Daniel Eth (yes, Eth is my actual last name) @daniel-eth.bsky.social · 27/03/2025

And if there is a software intelligence explosion, we could wind up with radically superhuman AI systems within months of ~fully automating AI R&D.

110

Daniel Eth (yes, Eth is my actual last name) @daniel-eth.bsky.social · 27/03/2025

OTOH, if the AI-improving-AI feedback loop overpowers diminishing returns, then AI progress continually accelerates, even if hardware is held constant. We dub this a “software intelligence explosion.”

110

Daniel Eth (yes, Eth is my actual last name) @daniel-eth.bsky.social · 27/03/2025

Diminishing returns to AI advancements may be steep – once the low-hanging fruit is picked, further improvements may be much harder. If diminishing returns are sufficiently steep, there won’t be an intelligence explosion.

110

Daniel Eth (yes, Eth is my actual last name) @daniel-eth.bsky.social · 27/03/2025

Once AI R&D is automated, will there be an intelligence explosion? This is determined by two opposing forces: 1) Positive feedback loop of AI improving AI 2) Diminishing returns to AI R&D (To be conservative, our analysis assumes that computing hardware is held fixed.)

120

Daniel Eth (yes, Eth is my actual last name) @daniel-eth.bsky.social · 27/03/2025

If AI R&D is fully automated, there will be a positive feedback loop: AI performs AI R&D -> AI progress -> better AI does AI R&D -> etc. Empirical evidence suggests this feedback loop could cause an intelligence explosion despite diminishing returns.

110

Daniel Eth (yes, Eth is my actual last name) @daniel-eth.bsky.social · 19/03/2025

There's debate over how long an AI system needs to act coherently to properly be described as AGI. My best guess is between 2 days worth of work and 6 weeks worth of work – hence, likely 2028-2030, given the extrapolation. Consistent with recent lab leader statements:

210

Daniel Eth (yes, Eth is my actual last name) @daniel-eth.bsky.social · 19/03/2025

Here's what happens if you extrapolate the figure further

382

Daniel Eth (yes, Eth is my actual last name) @daniel-eth.bsky.social · 10/03/2025

Anthropic researchers, trying to figure out why Manus is so good

0121

Daniel Eth (yes, Eth is my actual last name) @daniel-eth.bsky.social · 12/02/2025

Broke: preparing for AGI by buying TSMC & NVIDIA stock and shorting bonds Woke: preparing for AGI by getting Pakistani citizenship

2121

Daniel Eth (yes, Eth is my actual last name) @daniel-eth.bsky.social · 08/02/2025

At some point, some AI company is going to accidentally call their RSP a name that’s already been taken by another AI company. Imagine how embarrassing that’ll be!

1292

Daniel Eth (yes, Eth is my actual last name) @daniel-eth.bsky.social · 23/01/2025

Splitting a meal with some friends in 2007 Zimbabwe:

0170

Daniel Eth (yes, Eth is my actual last name) @daniel-eth.bsky.social · 08/01/2025

Good piece from Garrison Lovely arguing that, contra MSM claims, AI progress isn’t stalling but instead becoming invisible. Chatbots haven’t changed much, & that’s what the public sees. Meanwhile, AI has gotten much better at things like STEM research, which is a huge deal

1154

Daniel Eth (yes, Eth is my actual last name) @daniel-eth.bsky.social · 06/01/2025

When my grant applications to Open Phil and the Gates Foundation both get rejected:

0130