Ethan Mollick
@emollick
Followers
297K
Following
7K
Media
14K
Statuses
34K
Professor @Wharton studying AI, innovation & startups. Democratizing education using tech Book: https://t.co/CSmipbJ2jV Substack: https://t.co/UIBhxu4bgq
Philadelphia, PA
Joined May 2009
AI resources I work on that might be useful: My NY Times bestseller, Co-Intelligence (now in 19 languages!): https://t.co/pw21fSThMj The Generative AI Lab at Wharton (free prompts & research): https://t.co/hTOsEslqZf OneUsefulThing, my free newsletter: https://t.co/eQZg86qqwi
16
70
469
This doesn't mean that personas can't be useful - for example, they change how the AI answers questions, the format of output, and maybe other factors as well. However, you aren't going to make the AI suddenly better through role-play. Paper:
4
1
66
Get clear, healthy-looking nails with podiatrist recommended NONYX. NONYX is clinically proven for fungus damaged and discolored, thick, brittle, yellow and lifted nails. 90% of nails improve using NONYX. So effective its money-back guaranteed.
0
5
34
We tested one of the most common prompting techniques: giving the AI a persona to make it more accurate We found that telling the AI "you are a great physicist" doesn't make it significantly more accurate at answering physics questions, nor does "you are a lawyer" make it worse.
39
31
269
I asked Gemini for some similarly dark ideas and it came up with "A Modest Proposal: Tycoon Simulator" - which was absolutely too much.
4
0
17
I get that everyone has moved on the potential of a financial bubble as the next AI crisis (and maybe there is one, maybe not), but it would help to have a retrospective about how the various (very real) barriers to AI, from model collapse to pre-training scaling, were overcome.
4
3
89
There was a small flood of articles around GPT5 talking about how AI development has plateaued (partially based on experiences with the GPT5 router). Has there been any major articles with updates since it became clear that there was no such plateau? I still see lots of confusion
16
7
197
These three paragraphs from Kahneman in 2017 (pre-LLM) are something else - full of, as James says, "painful claims" that are grounded in a lifetime of research.
A painful claim from Danny Kahneman I can’t stop thinking about. “One implication is obvious. You should replace humans with algorithms whenever possible. Even when the algorithm does not do very well, humans do so poorly and are so noisy that, just by removing the noise, you can
50
128
1K
The worst literary game ideas, brought to life thanks to nano banana: Strunk and White as a dating simulator, Shirley Jackson’s The Lottery as an FPS, Ethan Frome as a racing game, The Yellow Wallpaper as a match-3 mobile room decoration game.
17
11
205
Expanding on these points: 1) Predictions that AI "will hit a wall" have not yet come true, though the future remains uncertain. However, lots of alternative approaches to continuing gains whether using LLMs or new architectures or hybrids are now being developed, and there is a
7
16
145
Summarizing 2025 in AI in a tweet 1) No sign of a slowdown in exponential pace of gains 2) Jaggedness remains the main issue of AI 3) Early days for deployment, but many companies reporting positive ROI 4) GenAI became an industry, with industry-level impacts 5) AI is still weird
35
87
582
AI created visual ads got 20% more clicks than ads created by human experts as part of their jobs... unless people knew the ads are AI-created, which lowers click-throughs to 31% less than human-made ads Importantly, the AI ads were selected by human experts from many AI options
14
41
363
Very neat paper combining art and economic history.
Wow, super interesting! "Colors of Growth" by Lars Boerner, Tim Reinicke, Samad Sarferaz, and Battista Severgnini. "We develop a novel approach to measuring long-run economic growth by exploiting systematic variation in the use of color in European paintings. Drawing
2
41
290
It does make me wonder about the underlying strategy of xAI - is it actively seeking to generate revenue to compete with other labs? Is it a SSI-like attempt to do a “straight shot” to super capable AI before commercializing? Is it just for synergies with other Musk enterprises?
4
1
37
For example, there is actually some odd stuff in their Grok 4.1 model card that would be helpful to understand better, including increasing sycophancy rates.
Interesting changes in Grok 4.1. Decreases in harmful responses but also increases in sycophancy and deception. It isn’t clear how to interpret the sycophancy score, but the MASK score for deception is quite high compared to big models. Sycophancy leads to higher LMArena scores
4
4
58
Compared to the other frontier labs, xAI releases very little information about their approaches, their benchmarking, or their safety procedures. Their short model cards are full of information that aren't well explained. This stuff does actually matter to corporate audiences.
2
3
60
xAI keeps producing very good models, but with little information or transparency about their approaches to AI overall (what is "truth-seeking"?), or the safeguards they have around altering models on-the-fly. Especially for enterprise use, more transparency could matter a lot.
22
18
208