Fraser Paine
@FraserPaine
Followers
116
Following
4K
Media
34
Statuses
638
Head of AI at 'a company', just trying to steer us toward a fully automated luxury future in space
Joined October 2011
Money is a sign of poverty, if you have to keep a track of resources you don't have enough to go around.
0
0
1
Bearish on AI doing all work any time soon because the people in SF making it have never had a real job.
0
0
1
There's about to be a Midjourney moment for 3D models. Its already here in Discords where people orchestrate Comfy.UI to use the best mesh generators, add PBR materials, and retopologise in one flow, but it's complex. This will be better and a single query within 3-6 months.
0
0
0
Coming up on a year since the last Generally Intelligent pod with @kanjun and there's still not a week that goes by I don't check if there's a new one 🥲.
2
0
2
RT @granawkins: > be me.> exponential curve supervisor .> get to work.> check-in on the curve.> it’s sigmoid.
0
206
0
Modern evals are pointless distractions. I don't care about multi-choice questions or models doing arithmetic. Until we standardized testing models using code interpreters and search engines and evaluate their answers long form questions its just irrelevant to real world usage.
0
0
0
We're entering the race condition era of AI development. Increasingly hard to forecast, despite the technical dynamics being better understood than ever.
0
0
0
Don't know if they're going to win but man, the @PrimeIntellect team have got taste.
0
0
5
Why is there no effort going into AI to automate data platform engineering?.Databricks are the only ones starting to take this seriously. The data platform is the foundation of any enterprise AI solution, yet there are 30+ SWE AI Agents and literally no serious DPE AI Agents.
0
0
0
I don't know about everyone else but I like my y-axis to not be "arbitrary", and for the x-axis to not round to the year. A literally useless plot. It's a shame that the popular anti-AI guy is an ignorant bully, we could be having much better discussions if he was competent.
Time to calm down about AI, at least about LLMs. A rough visualisation of the "diminishing returns" view of current progress, as explained by @GaryMarcus
0
0
0
So far I've been able to tap this sign three times in the past 3 months. Wonder if it'll hold true till Christmas.
GPT 4.5 and Claude 4 will be great, but Google are probably going to drop something in the next ~4 months that will take the lead and then they'll just remain dominant from there.
0
0
1
Well orchestrated AI can do a meaningful percent of knowledge work today, but it doesn't. Besides just pace of adoption and risk aversion, what other forces are limiting uptake?.
0
0
1
Deepseek quiet on twitter for about 7 weeks, dropped Deepseek Prover on Github 2 weeks ago with no announcement. I can only assume they're cooking something, I just pray its not a miss like Llama 4. .
0
0
1
This is still the most important concept in AI dev. Vibe coding is rightfully ridiculed, but there is insane potential if you provide detailed specs, code review, and test extensively. Not sure if Tiger Mom Coding is the right name, but current anti vibecoding sentiments are.
i do whatever the opposite of vibe coding is (with AI). i do tiger mom coding.
0
0
2
I spent a few hours discussing what an ideal future might look like with GPT 4.5, and then asked 4o img gen to show me a scene from that world. From there, I could pretty much just ask for another scene over and over again and see a dozen vignettes of utopia.
0
0
2
I just want to say for the record, I was completely wrong here. In real world usage, 4.5 is the best model available for clear, concise professional writing in reports and proposals, this is a common use case personally and I find it extremely valuable. Very sad to see it go.
GPT 4.5 shouldn't have been released imo, 30x the price of 4o, no improvement on any notable tasks. Weaker than Claude 3.7. Claim of good creative writing is not compelling for the cost. This was an 11 month old failed attempt at GPT5, their hand was forced by Anthropic and XAI.
0
0
2
Beyond AI just making things faster and easier, I've recently realised I'm making more things in general because the perceived effort is so much lower and I have less mental friction to just getting started. Is this accounted for in productivity forecasts for AI augmentation?.
0
0
1
We are halfway through the 20s, they are not roaring hard enough tbh.
0
0
2
GPT 4.5 shouldn't have been released imo, 30x the price of 4o, no improvement on any notable tasks. Weaker than Claude 3.7. Claim of good creative writing is not compelling for the cost. This was an 11 month old failed attempt at GPT5, their hand was forced by Anthropic and XAI.
0
0
6