iiinova — where your digital self keeps evolving

Konrad Kording @kordinglab.bsky.social · 04/09/2026

This is cool. But why don't people know how to pour coffee? www.biorxiv.org/content/10.6... the paper asks how to pour a coffee and the answer is "not like this!"

2100

Konrad Kording @kordinglab.bsky.social · 25/08/2026

And all of this, is better with better “training“ goals. Wherever truth is defined, push towards this truth (D). And where it is not defined, use LLMs to check answers ("self evaluation", (E))

180

Konrad Kording @kordinglab.bsky.social · 25/08/2026

A third idea (C) for making answers, combine many components into "Agents", : coordinate component large language models (LLMs), memory, databases, and tools

260

Konrad Kording @kordinglab.bsky.social · 25/08/2026

A second idea (B) for making answers, tool use. Call programs like databases, internet search, calculators etc.

160

Konrad Kording @kordinglab.bsky.social · 25/08/2026

A first idea (A) for making answers, inner speech. Produce words that are just like a scratchpad which the user never sees.

160

Konrad Kording @kordinglab.bsky.social · 25/08/2026

The thing that makes Neural networks so powerful is that there is a direct way (4) of improving them with examples and towards any definition of answer quality (training).

170

Konrad Kording @kordinglab.bsky.social · 25/08/2026

The real work is done by a core machine, the neural network which is used to generate one word at a time based on its inputs. The probability of each word depends on prior words. And words are chosen one by one according to those probabilities.

270

Konrad Kording @kordinglab.bsky.social · 25/08/2026

All that LLMs do is convert prompt words into answer words (1). And the old recipe does that with two goals, language imitation (2) and people's rating of answer quality (3). Answers are produced word by word using a neural network.

1100

Konrad Kording @kordinglab.bsky.social · 25/08/2026

We will use everyday language where possible and mark technical terms in purple. Key is that modern LLMs have a component of making answers (in blue, called "inference") and a component of improving answers (red, called "training").

1120

Konrad Kording @kordinglab.bsky.social · 25/08/2026

Thread. I have yet to see a proper explanation of how LLMs actually work that is targeted to an informed general audience. So here is an explainer that I put together along with Tim @countzerozzz (inspired by @CIFAR_News LMB discussions).

1723060

Konrad Kording @kordinglab.bsky.social · 12/05/2026

Here is the preregistration. Missing: Primary outcome; exact ROI definitions; model family; interaction term; prediction direction; preprocessing/exclusions; multiple-testing correction; classifier plan; confirmatory vs exploratory hierarchy.

081

Konrad Kording @kordinglab.bsky.social · 05/05/2026

A useful feature that, I think, all reference managers should provide - every (AI selected) relevant paper comes with a summary of how it matters to the project at hand (AI generated). Check it in planyourscience.com

151

Konrad Kording @kordinglab.bsky.social · 04/05/2026

Ok. Then this aspect of word-use is the only thing leading to the disagreement, it would seem. How do you fit your use to the Merriam Webster definition?

130

Konrad Kording @kordinglab.bsky.social · 04/05/2026

My app that helps students think more carefully through their projects now looks really pretty by 2026 standards (imho). Try it out at planyourscience.com

0135

Konrad Kording @kordinglab.bsky.social · 04/05/2026

From my talk in italy last summer (Neuroscience in data deluge age): docs.google.com/presentation... --- would be super curious on you two's general take on this

271

Konrad Kording @kordinglab.bsky.social · 28/04/2026

Here are the most popular neuromodulators (made you a new github repo): github.com/koerding/neu...

3165

Konrad Kording @kordinglab.bsky.social · 28/04/2026

After rereading the cool hidden brain paper (www.frontiersin.org/journals/sys...) I decided to make a new version. Code and everything available here: github.com/koerding/neu... Most popular: hippocampus. Why are diagonal band, medial geniculate nucleus, and interposed nucleus so unpopular?

46520

Konrad Kording @kordinglab.bsky.social · 27/04/2026

If you use planyourscience.com you can now automatically generate the text needed to preregister your study on @OSFramework

1289

Konrad Kording @kordinglab.bsky.social · 15/04/2026

Everyone knows this graph. I tried to clean it up/ make it modern. Attention some dataloss is unavoidable. github.com/koerding/dop...

5333

Konrad Kording @kordinglab.bsky.social · 31/03/2026

Others believe having the right amount of micro data would enable better AI. Or simulations of whole nervous systems. While everyone has their own goals, the value of compilers would help everyone.

100

Konrad Kording @kordinglab.bsky.social · 31/03/2026

What would it enable? Entirely depends on whom you ask. Single cell modeling needs parameters, which we currently can't measure. Many physiology questions require micro-parameters. Many disease related questions require micro-data.

100

Konrad Kording @kordinglab.bsky.social · 31/03/2026

For now there is a wide gulf between imaging giving us structure and physiology giving us function. We argue that we need to build a compiler that translates the former into the latter.

110

Konrad Kording @kordinglab.bsky.social · 31/03/2026

The time is ripe for compilers because imaging is getting better and cheaper at an unbelievable speed - voxels per dollars is doubling every roughly 1.5 year, at rapidly improving quality. 3d-imaging becomes attractive for neuroscience.

150

Konrad Kording @kordinglab.bsky.social · 31/03/2026

Preprint out arguing that we should build the techology to translate (compile) molecularly annotated connectomes into dynamics. I think this is incredibly important. arxiv.org/abs/2603.25713

45016

Konrad Kording @kordinglab.bsky.social · 06/03/2026

Wow. LLMs are starting to self-reflect on being LLMs doing pattern matching. From a discussion about one of my papers:

2110

Konrad Kording @kordinglab.bsky.social · 09/02/2026

Planyourscience.com now applauds the good changes you make - and pushes back on the less good one.

0102

Konrad Kording @kordinglab.bsky.social · 03/01/2026

Substack makes it super easy to test different titles against one another. I experimented for a while. I did learn something through testing with a range of titles over a few months. Whoever reads my substack posts does not care one bit what my posts are titled as.

Two titles. One gets 27.3 and the other 27.7 open rate.

1170

Konrad Kording @kordinglab.bsky.social · 27/12/2025

I think this was not very good LLM probing. See the screeenshots of me trying to help. Do you see how I think you could get dramatically more mileage out of LLMs by probing better?

110

Konrad Kording @kordinglab.bsky.social · 20/11/2025

In any case, this is me trying to provide Konrad style mentoring as an app. It won't be perfect. It is far too ambitious. Open to collaboration on this. However, it is not open source and won't be. Why not? Automating (even parts) of science could open doors to junk.

210

Konrad Kording @kordinglab.bsky.social · 20/11/2025

If you have already started a project (e.g. written an abstract) or you want to see how a given project should have been planned you can import from files:

110

Konrad Kording @kordinglab.bsky.social · 20/11/2025

Obviously, science is collaborative. So you can share your project with someone else and edit it jointly in real time, google docs style.

110

Konrad Kording @kordinglab.bsky.social · 20/11/2025

To give better feedback/ chat experience, the app also knows you if you give it your @orcid.org . Planning science avoids wasted time, enshrines scientific rigor, leads to better papers. And makes science more fun.

110

Konrad Kording @kordinglab.bsky.social · 20/11/2025

When asking for feedback, the app obtains literature and insights from Semantic Scholar, Open Alex, and Consensus NLP - this far improves AI performance at knowing the literature:

110

Konrad Kording @kordinglab.bsky.social · 20/11/2025

It is an important skill to learn how to write papers. But there are also a lot of standards for writing. So the app exports as outlines/ papers/ todo lists.

110

Konrad Kording @kordinglab.bsky.social · 20/11/2025

For unconstrained questions/ brainstorming there is a chat that has detailed knowledge of the process and the project and helps in planning (and will eventually propose changes to the plan):

110

Konrad Kording @kordinglab.bsky.social · 20/11/2025

AI provides feedback based on extensive criteria designed for the app (also shout out to @gunnarblohm.bsky.social and Brett Mensh for feedback for me writing this). This feedback aims to help improve a concrete part of the project. Very constrained:

120

Konrad Kording @kordinglab.bsky.social · 20/11/2025

The user interface is designed to make it easy to enter the relevant information - it has full templates for most science use cases. It also automatically saves on a database and has professional grade login via orcid, google, or github.

110

Konrad Kording @kordinglab.bsky.social · 20/11/2025

I think almost all scientific projects should be planned carefully. And I think an app can dramatically improve that. So I wrote an app for that (free for now, if you can fund this let me know). I tested it quite a bit (>8000 users in beta so far). try it: planyourscience.com

16121

Konrad Kording @kordinglab.bsky.social · 13/11/2025

AI is growing exponentially at a timescale of months while the economy is growing at a timescale of decades. How should we model the economy with rapidly increasing intelligence tech and slowly increasing physical tech? New AI+econ paper.

2238

Konrad Kording @kordinglab.bsky.social · 31/07/2025

Its a new term and a lot of us are preparing their lectures. I am a huge fan of online ways of teaching. And my student Joey Rudoler @jrudoler.bsky.social made a great online stats course. Shiny app and all. Modern take on stats. Thought some of you may appreciate.

1375

Konrad Kording @kordinglab.bsky.social · 09/07/2025

Do I misread this? pmc.ncbi.nlm.nih.gov/articles/PMC...

100

Konrad Kording @kordinglab.bsky.social · 07/07/2025

Every generation has been richer for a long time and, yet, people feel we live in worse times than the past. Are you all believing the future will tank? Mostly because of AI? Difference academia vs industry? Are my friends just drawn non-randomly?

680

Konrad Kording @kordinglab.bsky.social · 06/06/2025

Apparently people appreciate highly focused talks. I will do my best to comply in the future. 1 talk, 1 idea it is. Its a bit hard for us older folks as we probably have accumulated many ideas.

How many ideas should a scientific talk be about? 1 (56%), 3 (30%) As many as possible (10%).

5392

Konrad Kording @kordinglab.bsky.social · 16/05/2025

It’s also 5 min from home so I really care.

130

Konrad Kording @kordinglab.bsky.social · 16/05/2025

Philadelphia has a new top attraction!

1100

Konrad Kording @kordinglab.bsky.social · 04/05/2025

Wow. Twitter is a lot more concerned about neuroscience than me!

most people believe most papers do not contribute.

140

Konrad Kording @kordinglab.bsky.social · 26/03/2025

The transmitter post on embracing complexity in brains (through AI) just came out. I love embracing complexity. But I fear we have not yet figured out what we should do in this new complex space.

1151

Konrad Kording @kordinglab.bsky.social · 17/03/2025

Here is chatgpt with some more detail:

120

Konrad Kording @kordinglab.bsky.social · 12/03/2025

And one attempt at writing out the FEP axioms:

270