itissid

very cool. Congrats on the launch. first half of 2026 was the year when every one made such internal tools one way or another. I am happy you guys could build a product out of it.

The cool thing is even though this tool addresses a few, use case for reducing cognitive load, they happen so frequently that they add up.

I think they key with cognitive load is that the agents often produce a lot of trace/docs etc, in the end only a small % of the traces really are important to the final changes, because concise changes are typically very small and self contained.

One thing i see this becoming important with is maintaining internal tooling built using this. I maintain an internal docs system that helps me do designing before i build code, that docs system is completely vibe coded and i can add features to it very fast, but now its all grown up. A challenge then is can i understand just enough about a new proposed change to approve it? That is key to the doc system not becoming a burden in of itself, while maintaining a tight core feature set.

Classification models lend themselves to sparsity and explainability. The good ones are very simple and economic to run on a laptop. If someone told you before 2022 that a json classifier was a product you would have laughed it off.

To add to this the more difficult problems in classification done on scale have always been about collecting "good" -ve examples, enough data to calibrate on every confidence interval and debugging outliers. And those are solved on a case by case basis by the company pursuing its own peculiar version of the problem.

Am I the only one who thinks this is just all hype?

In CS speak very roughly this would mean something like disproving an algorithm by giving it a case that fails it. Right?

Yeah but a highly productive last two decades of math research from https://en.wikipedia.org/wiki/Polymath_Project has come from collaboration.

And yet I(and many others publicly on x) chucked Claude for codex back in June. Codex is my workhorse. The main reason I use codex is because of its language. I just got really tired of reading long weirdly worded prose that I had to fix with some skill(though some people like matt pocock and dex horothy have good ideas on this, buts it's just wasteful). This is triply bad for learning newer stuff because it goes up and down the abstraction layer on any topic like mad. One moment it would be explaining a high level detail and then cite contrasts and then point indirectly to an implementation detail as an example. It's idea of explaining more abstractly was also weird in a different way.

Fable was better but I don't have 1000$/day to spend on it.

Hi thanks for the insight. Do you see a role for Control Systems(i.e. ones analogus to Instrumentation engineering) playing a role to modulate certain parts of continual learning? One very important way we learn are lived experiences, it's like telling memory:this part is more important( for emotional or social utility values), pay attention. Good or bad lived experiences both count. I guess is that a path that practical research is considering?

The core or the problem is also what you are describing was bought up by Yuval Harari in his interview with the economist.

To paraphrase some of his lines: We make decisions using our emotions and our thoughts. What makes us different from the AI is that we can be afraid.

To portray a guy as ordering "beef bulgogi", in the same breath as "email this rocket design marketing", while it _might_ seem appealing and resolute, though oddly fast paced, seems pretty ignorant of the human _quality_ that make most practical decisions messy.

All of this will be besides the point. Here is what's gonna happen. The frontier labs are just gonna keep building powerful models. AGI or not, open models in a year will be as powerful as Fable and Astra — probably by using em — and at a very soon enough point after that some one (a state or a few dozen people) with a few 100 GPUs is going to launch an unconscionable attack(if they have not already) that's gonna do a lot of damage.

Please for the love of god, just sit in a room with the government and put some restrictions around AI use before it harms a lot of people. Like tell the government to impose a minimum spend on frontier lab AI's spend on cyber defense and building every country's capabilities. The post-training mask for "I am a good assistant" is going to become a very sad joke when many people literally lose everything.

When one writes jupyter notebooks for DS you are not writing python. If you ask 10 DSs explain to me what python's attribute lookup model is and why is it different from other OO languages like say Java or C++, they would not care about it. The only thing DSs care about is the rich DS Library support and fast speed of protoyping. To a DS using jupyter this is almost the same feedback loop as a type system at compile time.

Have you tried using `uv`'s newer tools? They help a lot e.g. with linting speed, lock management, package dependency separation, correct python version mgmt and no need to fudge with venv.

Really? https://www.destroyallsoftware.com/talks/wat

Does it have something like litestream to backit up for specific production usecases (i.e. a single webserver is enough and downtime of a few mins is tolerable)?

What is being made is "what are our core values?" argument. One does not need to be a mathematician to know how poorly this worked for large communities when incentives are misaligned...

If a subset of mathematicians, use AI to condense timelines focusing on goal 6.2 exclusively and make rapid progress and reach a proverbial inflection point — one where value proposition of the using this new normal is too enticing to give up — everyone will ask: "This thing is so awesome. Why should I care about your values?"

I found opus 4.8 too agreeable and too wordy(as opposed to codex) and too agreeable. If you are reading documents generating by it was too much. TBH. Fable did a bit better on this. Anyone seen a marked difference with opus 5 on this?

If you love em dashes and commas you will also love this https://archive.ph/0dxpW from the newyorker by Mary Norris.

I was never a big reader as a kid, but I fell in love with the New Yorker, much later in life, where I kind of adopted their idea of: "Once you understand them, you can read through long sentences very easily". That is the newyorker style of writing. One falls into two buckets, you love that kind of writing or you hate it. If its the latter, you should adopt em-dashes and commas to improve your writing style.

There is just so much that can go wrong with rocket launches if you do cursory reading about these things. A few random examples: The lowest stage cant just be "turned down" or modulated because they work by burning an inner solid rubber lining that's like an annular cylinder. The first three stages are roughly doing the job sequence: `lift off the ground -> reach target altitude -> reach target orbital velocity` and each stage is modulated for atmospheric pressure, gravity. Another limit is you can't really design it for a human in the middle(like a jet plane), so the rocket's computer needs to do everything, and when the rocket's pitch or yaw or roll(in a manner of speaking) go off beyond a nominal range its game over.

ISRO has a very good track record of launching rockets with solid state engines. I do wonder if almost all of their expertise was used for the first three stages(and they are not 3d printed)? And how much more difficult it is to make the last stage as compared to the first three, which as I understand, was the stage designed and tested by skyroot itself(manufactured by Wipro 3D).

Noob Question: There is a famous parlor trick with generative networks(I think it was GANs but it might be some kind of diffusion based network.), you start with a canvas and draw a stick figure of what you want and the generative network draws the rest of it.

Do AI platform companies actually pre-train networks to do the same for hand drawn artists?

Related question: If they do train them to do that, are there any that train people for the "reverse": learn how to draw with paper and pencil by showing techniques only i.e only the "what" but not the "how" ?

Ah so you extract decisions, actions and outcomes and you index and search over them?

Especially for indie users/devs and smaller teams. I built a part of this(the retriever) in < 4 hours https://github.com/itissid/wiki for replacing deepwiki.

I think the challenge is to teach how ranking works to people more effectively so that they can build it for themselves and host them on their own.

Like the other day someone who has worked in search explained to me why you would care about using learning-to-rank(LTR) technique to train your own feature vector weights on your data. My understanding is that weighted features work better(retreival wise) on textual data than plain BM-25 and vector embedding db indexing of text chunks of your data with minimal preprocessing. So if you have lots of conversations you can create a ton of features(like attributes of a conversation) from it and ones that matter more will rank higher. And you can use a regularization(like L1) to kill unimportant ones.

[EDIT]: IIUC, I think LTR is important because you likely want different features to matter more for different parts of your documents, e.g. what matters for codebase documentation is different from your personal journal.

I have a request: can this text be even more AI generated?

I think the main reason for this outrage is partly a confluence of factors some of which you pointed it.

But my money/theory is on what I call `alonliness` specifically due to lack of labor mobility: There is no good rapid transport connecting american interior towns to big cities and people don't want to move due to a variety of factors(real estate debt, cultural affinity to area). To be clear these people are alone but not necessary lonely(there is that going on too in other segments of the population). And once you don't understand or never had the opportunity to understand something(or rather only experience its bad side effects), you distrust it at best and fear and hate it at the worst.

If you ask people 1:1 IRL who live in communities where the main source of employment is not whatever Silicon valley businesses are. They will ask you questions or make comments like " 'those' jobs are not real American jobs!" or "Aren't you afraid of getting thrown on subway tracks in NYC?". These are actual questions i got from people there.

These give an indication of the disjointedness of the sets in the venn-diagram of of the socio-economic equality and what creates such psyche. I am not sure why some think its a PR/Comms problem for big tech.

Interesting. Making low latency correct tool calls correctly is pretty important in voice AI cascading models(STT LLM TTS). Realtime Models are still 2x the cost and there are only 2 providers openai and google that are in the race. For cost control it has to be cascading models

For llms Sadly the only model right now that fits the bill for LLM is GPT 4.1 and it’s standard in my stack because thinking models have unacceptable latency(>=1 sec) even though they are good at tool calling. The main issue with 4.1 is that it can make still mistakes and prompt prose has to be tuned quite a bit.

I wonder if any local models can be tuned to match the response time and tool calling while supporting many languages.

Still too narrow a take on what self help techniques are killable by ai. I also think self help as a bunch of life hacks and habits is precisely what’s wrong with the industry writ large. They are based on what sells in a capitalist system targeting the attention economy. Creating scarcity on supply side of attention(work longer) and demand side(addictive apps). Take Atomic Habits, one of the things it says is out of many is making the habit you want to form easy to do, e.g place sneakers and shorts next to your bed to make a habit to run in the morning. It presumes a lot about how exactly habits are actually internalized, retained and how one falls into or out of them. I’ve fallen in and out of such habits even after I did them for a while(months). Techniques are very ripe to disruption because people don’t quite understand or have time to understand or observe their own mental state, so hacks sell because you can do them(presumably) as a monkey would.

It’s not an electronic problem but an human first IRL interface problem. A shining example to the contrary is meditation practice like Vipassana. Saying you can kill that with AI is like saying “Gandalf is here and he explained to you the meaning of life and said now you don’t have to live or learn lessons anymore because you know I can always ask him”. Of course living the actual life is the whole point! It’s also why IRL experiences like classes and communities tend to work better when structured as lived experiences.

If this industry of self help books dies I won’t shed too many tears.

#TIL multiple myeloma went from being a death sentence in the 90's to quite manageable using a really old compound called thalidomide[1]. Though it sad that it was exploited for almost two decades to line the pockets of a dozen rich a*holes.

[1](https://www.propublica.org/podcast/revlimid-cancer-drugs-fda...)

I wonder if the government might consider reversing the circadian cycle and tell people to work in the night and sleep in the day. At this point its better because IIUC many animal's reproductive system shuts down at such high temperatures.

And I would not discount this govt considering it given that far less rational, crazier, things have happened in the past like for instance randomly cancelling high denomination cash notes.

Yep and the speaker is not half bad! It has good hardware. Its a shame they could not spin it off to anyone else(not for lack of trying) before they were "forced" to kill it because, basically, meta's stock was at 90$ at the end of 2022 and everyone was just spewing a derivative of the statement: "Focus on extremely short term profitability else your(heavily) stock compensated employees will all leave."

I've heard there is a way to know if your resume is getting blackholed by a AI resume screening process — sort of like a test to fix it up. Any one knows how that works?

> People generally limit the number of roofs to one, as they are expensive and important for keeping the outside out and the inside in.

What does this mean?

And why is it the worst place to put solar panels? Is this and America only phenomenon, cause in India people are installing them like hot cakes on the roof. What’s different about roofs here?

Has anyone gone back to doing code katas, code craft like exercises by hand? They help keep me grounded.

Also I feel like it’s fine to let AI write your code. I felt very much like the OP did. A couple of things help keep my sanity. one is that as developers I think our job has evolved to knowing what decision an AI makes is good and which one is bad, this can be code or design – but there is nowhere a developer(or for that matter a knowledge worker) can hide from ai. In this world you will be forced to communicate with them. Partly because as a community we have decided(for better or worst) that AI should bring non trivial amounts of productivity gains to software development.

The other one is something I want to validate which is for those of us who are mediocre at coding, it might be a gift because it would free up some time and thus mind space to consider what we are actually good at.

There is no difference between claude -p and typing into their "harness" in tmux. They want to make you pay for API which is subsidizing these plans.

They say they want to do it due to their percieved misuse(via automation) of `claude -p`, this I find laughable because they can always rate limit based on how fast a human types/STT's into a text box. Heck, they could provide TUI based extensions(something you vibe code in a few days today) to get better observability and DX for harness use, I would not mind using `claude` CLI directly, but they wont.

Forcing people to using their garbage harness — 0 observability, no extensibility and poor DX — is a showcase of how this fiasco is playing out: They refuse to improve their CLI and want to push everyone to use their API with multiple poorly rolled out policies.