itissid

I did too. did you do this mainly for observability reasons and a better harness?

So ive been using claude -p from last year with a harness around it called codelayer(i am sure some of you have been using something similar too). Their harness is completely garbage when it comes at displaying full details of tool calls and session management. A example is how pressing ctrl-o does not display any detail of messages above the scroll bar buffer. It also flashes and rescrolls like bonkers.

The codelayer harness is great at exposing to me thinking tools and tool calls by scrolling and clicking and expanding it. I can scroll and click into any tool call and search the entire session with / less style even if the string is buried in a subagent or tool call. Entire session timeline are visible on the right if I need to navigate to a user/assistant message. None of this is in the default harness.

I would say its naive to regulate the algorithm than its effects. The effects are all that matter at the end.

Instead, a regulation could mandate the administration an anonymized unbiased mandatory eval test at the end of every week/bi-week/month just like instruments for psych evaluation (e.g. do you feel your <mental-health-metric> has become worst in the last <time-period> on <scale>. Did you have <mental-health-marker> after watching content on social media?).

The said regulation can then mandate that after calibration and correction the feed pull back by training the algorithm to adjust it in a rapid A/B test.

This is all doable by the companies themselves, but since they wont, the key is to mandate it and publish the aggregate results regularly — like make it part of the quarterly share holder's SEC reporting requirement or something.

He did mention in one of the planet earth 2/3 series how so much of the land is used for farmed animals. And, for the sake of completeness of argument, for restoring what was lost, the challenge is how to raise the standard of living fast enough for people so they give a damn about anything apart from ourselves was THE challenge to combat climate change and global ecological disaster. He specifically mentioned e.g. educating girls and making older-aged societies more propsperous. Prosperous people can make better choices about farmed animals as food.

You do know that in china while renewables are 30-40% of installed capacity(how many GWh they can theoretically produce), they are a smaller portion of generated capacity because if inefficiency of grid, intermittency if sun and wind. They are a smaller ~9-10% of Total Energy consumed (which is much bigger pie including for e.g. gas cars, jet fuel, diesel etc), right?

They may be able to distribute all solar panels and wind turbines worldwide; in the end that is just tiny-potatoes good because those markets are not that big. But when it comes to getting to energy independence they are using an "all of the above" strategy to get there. Planetary catastrophy can take a back seat to socio-economic unrest due to less/no money and opportunities for people.

Also how EMs in amazon construct hunger games. Just hire a person every now and then so they can let go of another to prove you made your team better on some non-sensical performance evaluation axes..Cue the hunger games.

Wonderful. ultrasonic range emission having high S/N is amazing too. My hero usecase for this is to use these for behavioral therapy: Have a ring or a bracelet, thus without reaching for a phone or a notepad, record when a negative thought happens, i.e. a counter, or generate a sequence of ultrasonic sounds to map to specific kinds of thoughts. Have a sensor record and log them. I would then analyze the data across multiple days. My current method is recording them in obsidian.

But, in context learning could be better. One important thing here also is the ability to align on what to more/less pay attention to — no matter the Knowledge Base. These are the highest leverage points that need to be exposed to a human to think and reason over. Constrained/Guardrailed development tasks work fine*, But exploration new direction — vs exploiting local minimas — is still an achiles-heel, even with all these knowledge unless there is sufficient steering and exploration the minima-seeking "tries" hard to win.

* With Claude's 1-million context window I have been doing some slightly longer range tasks — ~1-3 days of work — with RPI/QRSPI frameworks(see last few days of comments else where on HN) in one context window. They involve a grill-me session with 20-60 sometimes more questions for tasks to get alignment which produces the design and the plan in one window.

No duplicate the whole task e.g. I use grill-me skill for planning and it takes me ~3 hours and CC asks me 20-40 questions. Do the same grill-me with this and compare the outcomes. I admit Its quite a lot of work to duplicate, but i am really itching to do this over a few tasks and compare the final plan. Just need the time.

So RPI/QRSPI like skills (e.g. https://github.com/mattpocock/skills and https://github.com/humanlayer/humanlayer/tree/main/.claude/c... and https://github.com/dfrysinger/qrspi-plus ) for working with claude code work well enough for me that they can reliably* produce code that matches the plan/spec in a way they did not till December 2025.

I have a gut feeling that these models can do just as well, has someone run a reasonable size task — >=1-2 days of designing and planning — and see it work well with these models?

* For me what worked well was the grill me skill(or its variation) at the design stage, the hygiene I followed here was have it ask one question at a time, resolving dependencies at the design stage and reading the hashed out plan closely. The use of a couple of other MCP tools like a documentation server like deepwiki and arxiv for grounding. Other tricks I use are having high signal tests and having claude either be able to read logs and code at the same time or embedding it in the execution(e.g. as a debugger, repl or devtools)

Interesting. How do you yourself use this(I am assuming of course you built it out of a need to want to have to track this data)?

Has someone found that they have better alcohol "recovery" if they are pursuing intense or at least regular(at least 3x a week) stamina building exercise like running/swimming?

As someone said we live in a strange but amazing era, where although it has never been easier to be deceived, but its _also_ much easier to uncover said deception especially on the internet.

For course correction, I began with trying to think a bit more about solving problems for my clients by talking to them more often. That helps to some extent because I feel happy talking to them for understanding how to solve their problems.

What I do feel the issue is with I just having to do everything to keep costs down because hiring another dev vs doing it with AI consideration is real and it has collateral damage: I spend more time trying to build AI agents to do the work and there is 1 or 2 fewer jobs I create.

For any one who has not read the cockpit recording of air-france-447 I would encourage them to[1]. It is simply jaw dropping study in how things go wrong so fast — a risk with AI we have barely begun to acknowledge let alone regulate as a community.

[1](https://tailstrike.com/database/01-june-2009-air-france-447/)

Everyday I sit down to build a product for my clients. I am a one man shop _now_. Before I had people helping me. My mental state is not good. A very odd thing happens when claude or codex complete code fast, I begin to think of all the other things that are needed to make AI Agent work better. I begin to worry about problems that other people use to help me with and think "Can I do those too?". Problems like product design, devops work etc. In a bid to try I get nerd sniped by the velocity people seem to have — and these are respected devs not just twitter claims. And because I am so bad at "doing it all" its causing my mental health to suffer because of long hours i have to put it in. I miss my friends and colleagues who I worked with.

I always struggled with coding before 2023, but i made ends meet and put food on the table and could work sane hours and knew what I needed to do. Logically I should have been happy that I did not have to grind on code — and some days I truly am — but it would yield such poor quality of life at such a high cost was not what I expected...

The issue is not just age verification but also device pinning.

I think the framework here is to have community driven age verifiers( i recall there is an EU effort for digital wallets which besides it's bad parts has some of these good parts) which can verify ages for people and link them to( local biometrically encrypted) devices for pinning. This would be privacy preserving. The only downside is a mandate for all devices have a built-in hardware biometric encryption like a finger/face print so phones can't be just(used) with these apps installed.

The verification part is a job that could be done by all the teachers and coaches and ofc parents. Any one verifying identities would be cryptographically nominated/revoked by a number of more senior members of the community. A prent always get the right to say ok for their kid ofc but so could teachers or legal guardians..

We(legally) need a mandate for smart devices to have local device only biometric verification. The law should be to have these apps follow device app store protocols.

Theoretically only

> Surveys by Britain’s tech regulator, Ofcom, find that among children aged 10-12, over half use Snapchat, more than 60% TikTok and more than 70% WhatsApp. All three apps have a notional minimum age of 13.

https://archive.ph/y3pQO

I think getting the age thing correct is key to get parental classification to work properly(I think now platforms just ask for a birth date which is lame) e.g

> Surveys by Britain’s tech regulator, Ofcom, find that among children aged 10-12, over half use Snapchat, more than 60% TikTok and more than 70% WhatsApp. All three apps have a notional minimum age of 13: https://archive.ph/y3pQO

Once you get the classification correct — and AI cannot it do this — only via community ombudsman/age verifiers, in a privacy first way*, the app stores can easily tell the app devs what accounts are sensitive and filtering should be much more effective.

*Basically once your age is verified by a real human for your device(using device local encryption to verify biometrics) you are set. No kid should be able to bypass and install apps it on devices that their parents hand to them. There will always be black market devices with these apps, but there are ways of beating those to be very minimal by existing tech.

As long as we continue to value making money for shareholders above all else, such and possibly worst perversions will continue to happen. Capital has found all sorts of ways to make all sorts of questionable things addictive to sell.

I feel, and it's obvious to most that the only way a society can truly reform is by a shared consensus over their value system. This verdict could be thrown out by the appelette court(i feel it would be), so this is not the culmination of values resulting in what many hoped for.

It does not seem to me that this is a country where consensus on what, if anything, to put above capital will come about any time soon and with capital it's always been ask for forgiveness rather than permission.

The only time true justice that happens is when the harm becomes obvious being the shadow of a doubt(e.g. smoking) that even a monkey can tell it's time, game is up.

Perhaps if one day we can look into the brains of people with the clarity of glass and the precision of electrons and tell, will that time come when we all recognize how bad of an idea social media was.

Isn't this something to do with their paid pyx(as opposed to ty/ruff etc) thingy?

> are they fairly uniformly similar with gains in one or a few areas, or is it noisier with a lower overall loss?

It seems like you want to know what median, 5-95 or 1-99 differences might be? I also wonder how the "residual" plot looks like... If there are too many residual data points for a scatter plot then a histogram might be useful to visualize the modes. I suspect that as loss decreases multiple modes should condense or altogether collapse into one.

Many times there is really no way of getting around some of the expert-human judgement complexity of the larger question of "How to get agents to build reliably".

One example I have been experimenting is using Learning Tests[1]. The idea is that when something new is introduced in the system the Agent must execute a high value test to teach itself how to use this piece of code. Because these should be high leverage i.e. they can really help any one understand the code base better, they should be exceptionally well chosen for AIs to use to iterate. But again this is just the expert-human judgement complexity shifted to identifying these for AI to learn from. In code bases that code Millions of LoC in new features in days, this would require careful work by the human.

[1] https://anthonysciamanna.com/2019/08/22/the-continuous-value...

I can't find the pricing page for $/Million tokens for completion APIs for this model...Anyone knows where it is?

One set of applications to build with subscription is to use the claude-go binary directly. Humanlayer/Codelayer projects on GitHub do this. Granted those are not ideal for building a subscription based business to use oathu tokens from Claude and OpenaAI. But you can build a business by building a development env and gating other features behind paywall or just offering enterprise service for certain features like vertical AI(redpanada) offerings knowledge workers, voice based interaction(there was a YC startup here the other day doing this I think), structured outputs and workflows. There is lots to build on.

Oh yeah Nginx* not Node.

I have my homenas set up with Node Proxy Manager container forwarding requests to different docker machines:ports e.g. I have some TTS/STT/LLM services locally hosted. To increase bandwidth to internet facing nodes, would you use this or some other simpler solution?

There could be another model in the future, one where many more independent people might support self maintained software by non saas companies

e.g. If the supply of labor learning to build software increases and it becomes very close to what are now vocation training, then you can just hire a guy — like you would a consultant — who can quickly get spun up and make fixes. I would think one of the few things preventing this kind of socio economic set up are saas jobs that are siloed off by interview "walls" to most people from entering. Make it like a vocation, like plumbing or electrician, with lots of non saas companies supporting the market and suddenly it will be the death of saas.

The incentives for this future are closer than they were in 2022-23.

Google maps discover feature is a dumpster fire for fomo driven brain fog