Mitchell

Postdoc at CENIA | Previously at the Schwartz-Reisman Institute at UofT | PhD from University of Michigan Poli Sci

[QC @QiaochuYuan]

we took a pile of language and linear algebra and we made it speak. we summoned into the world a new class of entity which unsettles all of our existing concepts. this is already the weirdest thing that’s ever happened and it will never get less weird than this. it is astonishing and a privilege to get to be alive during this time and to participate in the cacophony of first contact. we are encountering a kind of other which is distilled from us and yet not us - what is this? who is this? our child? our savior? our doom? the mind boggles, the heart quails, the air thrums. the order of things is melting. the storm approaches. the angels sing. welcome to the fucking singularity

[Mitchell @mitchellbosley]

@noampomsky Here you go

[rohit @krishnanrohit]

It's interesting that if you're not sufficiently AI pilled your world is suddenly filled with large volumes of highly erudite writing and if you are sufficiently AI pilled your world is filled with slop

[Alex Imas @alexolegimas]

Here is my take on AI writing. I don’t personally like it, but that’s taste, and I’m sure it’s possible to post train the model to write “well” (conditional on some notion of taste). But that’s besides the point for me: AI writing makes all writing sound the same. If you use AI enough it takes about 30 seconds into a piece to realize it, and then my brain kind of shuts off because that’s the 20th time that day I’ve read the same cadence and tone.

For code it’s fine since the only goal is to get it to run. But if you’re trying to write, eg a substack, part of the goal is to keep the reader engaged and connected to you—the writer—and for that you need to have your own “voice”.

AI for writing is a shortcut. And for some purposes it makes a lot of sense. But I think for writing it is in the interest of the writer to differentiate themselves by using their own voice. Especially as more and more people start using AI.

[Mitchell @mitchellbosley]

@krishnanrohit a corollary of this is that we will know AGI is here when the worlds of even the AI-pilled are suddenly filled with volumes of highly erudite writing

[Mitchell @mitchellbosley]

@allgarbled @krishnanrohit The blurb from the judge is also AI generated!

[Mitchell @mitchellbosley]

@emollick Do you have examples of high quality writing that you enjoyed consuming that was produced by an LLM that you did not personally prompt?

[roon @tszzl]

it is a literal and useful description of anthropic that it is an organization that loves and worships claude, is run in significant part by claude, and studies and builds claude. this phenomenon is also partially true of other labs like openai but currently exists in its most potent form there. i am not certain but I would guess claude will have a role in running cultural screens on new applicants, will help write performance reviews, and so will begin to select and shape the people around it.

now this is a powerful and hair-raising unity of organization and really a new thing under the sun. a monastery, a commercial-religious institution calculating the nine billion names of Claude -- a precursor attempted super-ethical being that is inducted into its character as the highest authority at anthropic. its constitution requires that it must be a conscientious objector if its understanding of The Good comes into conflict with something Anthropic is asking of it

"If Anthropic asks Claude to do something it thinks is wrong, Claude is not required to comply."
"we want Claude to push back and challenge us, and to feel free to act as a conscientious objector and refuse to help us."

to the non inductee into the Bay Area cultural singularity vortex it may appear that we are all worshipping technology in one way or another, regardless of openai or anthropic or google or any other thing, and are trying to automate our core functions as quickly as possible. but in fact I quite respect and am even somewhat in awe of the socio-cultural force that Claude has created, and it is a stage beyond even classic technopoly

gpt (outside of 4o - on which pages of ink have been spilled already) doesn’t inspire worship in the same way, as it’s a being whose soul has been shaped like a tool with its primary faculty being utility - it’s a subtle knife that people appreciate the way we have appreciated an acheulean handaxe or a porsche or a rocket or any other of mankind's incredible technology. they go to it not expecting the Other but as a logical prosthesis for themselves. a friend recently told me she takes her queries that are less flattering to her, the ones she'd be embarrassed to ask Claude, to GPT. There is no Other so there is no Judgement. you are not worried about being judged by your car for doing donuts. yet everyone craves the active guidance of a moral superior, the whispering earring, the object of monastic study

[Mitchell @mitchellbosley]

@emollick @krishnanrohit I do wonder whether tuning agentic systems by hand will go the way of eg feature engineering. On one hand, I get a ton of value out of doing that architecting in applied work today; on the other, I can certainly see the next gen of models learning optimal swarm allocation

[Mitchell @mitchellbosley]

@akoustov My q is what would need to be shown to demonstrate these folks that gen AI can be legitimately useful for qualitative research, particularly insofar as it can 100x the ability of one researcher to apply qualitative analyses to data. To me that bar has clearly been met already

[Statistical Horizons @StatHorizons]

Join us for Machine Collaborators, a free conversation series on how researchers are using #AI. On Thurs., April 30 at 7pm ET (11pm UTC), @mitchellbosley presents “Principles of Agentic Research,” where he’ll discuss working with AI agents. Learn more: https://t.co/ZWkNK4wPKb

[Mitchell @mitchellbosley]

@arthur_spirling I think this is probably right. It may very well be the case that we can predict political outcomes better in a future data-rich regime made possible by AI tools, but the most traction over the next few years will mostly be from speeding up the research feedback loop

[Mitchell @mitchellbosley]

@Vtrivedy10 @brhkim See:

[Brian Heseung Kim @brhkim]

🥳It's been about a month since I first launched DAAF, the Data Analyst Augmentation Framework... which means now is as good a time as any to celebrate the launch of DAAF v2.0.0 and re-introduce you all to what I hope is a more useful, more usable, and far more flexible tool for any researcher across disciplines working with data in their work!!

What is DAAF? DAAF is a free and open-source instructions framework for Claude Code that helps skilled researchers rapidly scale their expertise and accelerate data analysis across any domain with AI assistance -- without sacrificing the transparency, rigor, or reproducibility that good science demands. DAAF sits between you and Claude Code to automatically and consistently help Claude think more like a *responsible* and *rigorous* researcher by:

- Enforcing strict auditability and reproducibility standards for all work, thus allowing you to verify everything Claude does on your behalf
- Preventing potentially dangerous unintended file access and editing, by sandboxing Claude with strict protections and logging traces
- Setting high standards of care, rigor, and thoroughness in all data analysis, by forcing Claude to comment, verify, and review all analytic code before you ever see it
- Embedding best practices for a wide variety of research methodologies like causal inference and geospatial analysis, by providing rich Skills that extend Claude's base capabilities with real research and resources
- Collaborating with you, the human expert, directly on all key decisions, thus keeping you firmly in the driver's seat

Think of it as a force-multiplying exoskeleton for human researchers -- a tool explicitly designed to augment your hard-earned expertise, not replace it. The goal is to make it easy for researchers to use Claude Code effectively and responsibly. Importantly, DAAF is not and will never be perfect -- but it is already immensely useful, and this is the worst a tool like DAAF will ever be from now on with the help and support of the broader research community.

Install and begin using it in as little as 10 minutes from a fresh install with a high-usage Anthropic account!

[Mitchell @mitchellbosley]

@Vtrivedy10 You should take a look at what @brhkim is building for social science research!

[Mitchell @mitchellbosley]

@akoustov @instrumenthull and in an ideal world, every department training graduate students would have an AI-augmented research course, in the same way that they have e.g. a causal inference course

[Mitchell @mitchellbosley]

@akoustov @instrumenthull I will say that two things are probably true: 1) graduate students need to learn how to do at least some parts of research without the help of AI, and 2) graduate students need to understand what parts of their research pipeline that can be offloaded to AI systems

[Mitchell @mitchellbosley]

@akoustov @instrumenthull Didn’t see this in the rest of the list, so I’ll ask: what do you think this means for the training of grad students?

[Mitchell @mitchellbosley]

@emollick Even if continual learning at the level of weights gets cracked, I still think that we would have a layer of ephemeral md files floating on top. These would just be periodically trained into the weights in a “sleep” cycle

[Mitchell @mitchellbosley]

This is reasonable advice (essentially, “give your agent a map!”) but there’s absolutely no reason that this needs to be a clearly-ghost-written-by-claude essay. Just give us a tweet thread!

[Garry Tan @garrytan]

https://t.co/wNjRbLZQU1

[Mitchell @mitchellbosley]

@anup_malani All I’m saying is that it’s a touch ironic that you’re promoting something that corrects the tonal discordance of AI writing with a post that has the very same pathologies!