Christian Hendriksen

Associate Professor at Copenhagen Business School. Author of "AI på Arbejde"

[Christian Hendriksen @chehendriksen]

I'm sharing my practical guide to ChatGPT and Bing for social science and management studies.

This document part of the effort to democratize AI capabilities in university settings.

https://t.co/mgF8tZawEM

🧵 1/8

[Christian Hendriksen @chehendriksen]

@ChrissGPT This is the first time I've seen GPT-6 rumored for August. Any other signals on this? I would presume much later than August.

[Christian Hendriksen @chehendriksen]

Positively surprised at how much Fable I have on the Max (5x) plan. This is usage after one entire task I had saved for Fable.

[Christian Hendriksen @chehendriksen]

@deredleritt3r @peterrhague @joshgans 5.5 does a lot of stuff well, but there's some creativity or comprehensiveness that Fable displays which I don't find even in 5.5 Pro. Fable holds complex concepts and bind them together way more comprehensively than the GPT models. It's absolutely fascinating.

[Christian Hendriksen @chehendriksen]

@deredleritt3r @peterrhague I really don't think 5.5 is equal to Fable on research-grade social science. I think @joshgans agrees

[Christian Hendriksen @chehendriksen]

@MarcWarner10 Speaking as a university professor, there's a lot of 2) in 1)

[Christian Hendriksen @chehendriksen]

@Miles_Brundage 1) there's no single entity that can direct resources to ambitious projects, the Commission is too dependent on national budgets. 2) European policymakers are allergic to strong bets on technology that seems too sci-fi.

[dan @irl_danB]

imagine showing this to someone in 2022

Anthropic, newly formed => “worlds most valuable private companies”

negotiating with… Trump Admin (lol!)

export controls… on models!

“briefly the most powerful AI available to consumers” (!)

Mythos, Fable… (they called them what?!)

2022! this is not that long ago! this timeline has unfolded entirely in the span of my 4yo’s life!

[Christian Hendriksen @chehendriksen]

@Afinetheorem I think the intended audience does not recognize AI writing, so as ever, the author has an incentive to use AI for writing.

[Christian Hendriksen @chehendriksen]

@kimmonismus I'd wager 5.6 is much more intelligent per dollar (as @polynoamial has implied that the 5.x series is focused on) but also that Fable will remain better for certain tasks that require taste. I am not sure all of Fable's qualities can be reached simply with more test time compute.

[Christian Hendriksen @chehendriksen]

I'm doing a treasure hunt for my son's school and I had a bunch of pictures I wanted to add dinosaurs to for the clues. So I explained the problem to Codex, who promptly whipped out a custom-made tool where I could designate where the dino should be in each picture before handing back to Codex for image generation.

In some ways, the current reality feels quite sci-fi-like, but the net result in this process, like many other processes, is that I can spend more time out touching grass with my kids because I don't have to spend as much time in front of the screen.

[Christian Hendriksen @chehendriksen]

The two latest posts by @JohnJumperSci tell an interesting story.

[John Jumper @JohnJumperSci]

A bit of news: After nearly 9 years, I have decided to leave Google DeepMind and join Anthropic (after taking some time to recharge). I am incredibly grateful for my time at GDM. @demishassabis took a real chance letting me lead the AlphaFold team just six months after finishing my PhD, and the entire GDM team taught me so much about how to do great science. GDM is a special place, and I’ll still be excited to hear about what amazing things they discover next.

[Christian Hendriksen @chehendriksen]

@reach_vb I've been getting an error the past few weeks when running goals where the model seems to forget it can compact. Pretty annoying - what can be done? 5.5 Xtra High.

[Christian Hendriksen @chehendriksen]

@AndyMasley I'm also betting against this. It might become one of the most important issues, but not *the* most important one.

[Christian Hendriksen @chehendriksen]

Downstream effects of the Fable 5 situation is hitting Codex.

[Christian Hendriksen @chehendriksen]

Today, we are all Europoorians.

[Christian Hendriksen @chehendriksen]

@jasondeanlee It's decidedly better for complex writing and frontend. 5.5 Pro is not as good, and even when it comes close, it's quite jarring I have to access it in chat and not Codex.

[Christian Hendriksen @chehendriksen]

@justsomeoneDK My early impressions from Fable is that it requires significantly less guidance and hand-holding, but remains to be seen how this will pan out for papers.

[Christian Hendriksen @chehendriksen]

I think this is an interesting report by Ethan, but for the academics reading along, consider the example with the academic paper.

The first test I did with Fable was also to write a full paper, and it was really good. Probably better than all papers I've reviewed the past four years, perhaps. And now we are in a world where Fable can, to a single prompt, write out something that is probably better than many first drafts of human-made academic papers in social science.

We are in for a really fundamental discussion about what function academic articles serve, and although this discussion is not new, and we have had early indications of this by people like @joshgans and @ahall_research, I'm not sure academics (at least in social science) have internalized how good frontier models are getting.

[Ethan Mollick @emollick]

I've had access to Fable for a bit. A genuine jump in capability, I could feed it a 15 page design document for a project and it would work for 9+ hours and deliver terrific results.

But working with it is weird & weirder is coming

Lots of examples: https://t.co/HptkYunBzr

[Christian Hendriksen @chehendriksen]

@justsomeoneDK Probably not, but then again, I would be surprised if even 1/10 of the studies to Org Science are done with a frontier model right now (5.5 Pro or Fable)

[Christian Hendriksen @chehendriksen]

@justsomeoneDK Ah yes, you're right! All right, two prompts then.

On my own Fable run, I had a skill enabled that synthesizes best practice on academic writing, so I didn't need to give it more direction.