farhanhubble

This is cool and I understand the RAM emulation with gated attention but how does the CPU emulation wor? You add layers that are then trained to predict outputs while interpreting RISC instructions?

That's interesting. So, filling in details on demand with something like Gaussian splats and then backfilling them later on could save a lot of "wasted" time.

I kind of expected this. Fine-tuned models, among other tools, should be used by LLMs to interpret nuances in images. Just as I wouldn't trust LLMs to interpret a complex mathematical formula without it having access to Lean and other methods to check its reasoning.

Why was anyone using certificates from a little known company with no cybersecurity muscle?

I first discovered NuShell in 2023 and used it on and off for my data analysis work. For example, if I needed to check what unique values a particular field in a, deeply nested, JSON took on. With some practice it was far reliable than weeding out errors from bash scripts.

Although LLMs take most of that pain away, NuShell may still have a place. Xonsh: https://github.com/xonsh/xonsh is another very powerful shell but with a different design philosophy.

I know it's open sourced, but I'd love for the author to launch a Kickstarter product.

The way AI has consolidated power away from the hands of a few hundred large organization to a few dozen needs to be attacked. We need to build peer distillation networks. We need to be able to train frontier models in the open and design hardware for open inference.

As much as I detested the idea of Leetcoding the reason they became the de-facto for hiring was two fold:

- Software development for the most part did not require any particular specialization but it paid really well because of the explosive growth of the internet so there was an oversupply of talent.

- Most companies had no idea what they wanted from hires and how to test them so started copycat-ing puzzle based and algorithm and data science questions from the days when programming was mostly numerical computing

It devolved into a race to the bottom as more and more people lured by the 300k packages lined up and companies well aware of "No Silver Bullet" threw these people at increasingly menial tasks.

Companies at Leetcode and Hackerrank have no relevance now that you can ask an AI agent to do a hyper specialised candidate selection based on the role, with most roles now being redundant every six months.

Far from a breakthrough but being able to talk to embedded medical devices on the fly made me excited. The real test will be hardware debugging.

I once read The Bugs We Have to Kill: https://www.usenix.org/publications/login/aug15/bratus and one particular thing that has stuck with me forever:

“Any sufficiently complex input format is indistinguishable from bytecode; the code receiving it is indistinguishable from a vir- tual machine.”

Lisp may not have the ecosystem to compete with Python or JS. On a tangent though, I am in the middle of a mini experiment with Codex that stemmed from trying to understand the beauty of SectorLISP and is now headed towards: "What would it take to build a toy OS for RISC V using successively richer evaluators?".

Both pure vegan and pure meat camps claim to have scientific evidence that it works, which make me question if change is what the body responds to. From personal experience, eating less often and fasting does seem to improve things like brain fog, indigestion etc.

Just a few days ago I was pondering why an Emacs like environment that provides such powerful tools to examine and manipulate texts is not being used more prominently than say VSCode where you need to be build many features or use plugins or rely on system utilities.

I made the same point to an academician who was dejected about the prospects of future PhDs and Postdocs. My personal belief is that AI opens the gate for more unconventional and interdisciplinary research.

I'm reading all these accounts of people who were considered "smart but lazy", who were bright but not disciplined or driven to participate in things that gets grades. As a kid I had a near perfect memory and limitless curiosity. I would sit with all the textbooks before the academic year started, flip through them and feel excited. I also loved doing my own experiments, sometimes with near-perilous outcomes too. I was way ahead of the curve and did great academically too until high school.

Over time I lost my golden memory and started finding it hard to remember exact details and even vocabulary as easily as normal people do. The slowness and the academic rut also stifled my natural growth. I forced myself through a bachelor's in computer science. I loved every topic and went diving deep into many of them and I could explain them better than my professors, but I had no desire to pass silly exams and prove myself. I passed with a CGPA 8.4/10 and entered a workforce that followed the same rut. Silly, irrelevant interviews that force you to "prepare" for months for them. I did it early on to get my first job but never indulged in it since.

I have had a fulfilling career so far, working on interesting problems with real outcomes. I switched to data science. I dabbled on both sides of the industry; the research papers obsessed R&D and the whatever-shit-sticks side of software development and found mediocrity and sleaze on both sides. I never climbed the corporate ladder and never got close to getting loaded.

I think for smart people happiness comes from chasing exciting and meaningful goals, but the realities of life keep them chasing other more mediocre things and that takes a toll.

Personally, I have stopped looking to my work for intellectual satisfaction or growth or safety or camaraderie. It's a purely transactional relationship. I work sincerely and I get paid. I give them inputs on what can be improved and what makes more sense, but I don't go overboard and make any efforts beyond my paycheck.

I let my true self run amok in exploring things I do on my own, in applying the scientific and analytical mindset to everything from gardening to sewing to building better learning environments. If some of it brings a windfall someday, I'll be marginally happier.

I've not used Open Router but I'm wondering what are the reasons to route LLM queries at large scale?

Isn't the trend increasingly to let an orshestrator agent decide which model to use in terms of problem difficulty vs. size and thinking budget?

Once you fix those parameters there's little to optimize over, no? I mean you want to find the cheapest one across providers but that's not a recurring decision.

Does this type of atomic engineering have the potential for creating new types of materials, like polymers?

Regulation is necessary but it won't be sufficient. In India they have the do not call registry but despite that I get twenty odd spam calls a day whenever my number is active.

Personally I've benefited 100x more by using Truecaller but with time the app has been enshittified with weird features. That and my migration to iOS which has questionable support for Truecaller has made me the target of spammers again.

Furthermore many telemarketers are now using Whatsapp bypassing such regulations.

Hardware is hard and I applaud every effort to make it easier for tinkerers to build atop a platform. I've been away from it for a long time but how does the RPi system stack up in comparison? With all the hats available and the variety of cases, it shouldn't be too difficult to match the aesthetics, power consumption and exploit all the pluggable peripherals. What are the blind spots?

If anyone wants to take a peek into the Unicode codepoint, here's an indispensable tool: https://r12a.github.io/app-conversion/index.html

I started my software career maintaining and forking Unicode parsers. Arabic, Hindi, Chinese and Thai among many other complex languages. It was great fun and it helped me get a deep understanding of how complex writing was and appreciate the beauty of being able to reduce this complexity down to data structures and functions.

Absolutely cool demo. I used to love this feature in SwiftKey. It still works but after their acquisition by Microsoft they've gone downhill. I've since switched to iOS but would love to try Futo out when it's available for iOS

Easily 99 percent on most tasks. As an example, for a Python project with a dozen modules and ~50 files,a simple instruction like "Design a config file backed by Pydantic to store the project's settings. Keep the models modular" sets up nested Pydantic models, moves the settings to sensibly named JSON fields and updates the code to use Pydantic classes everywhere. Takes a few minutes maybe. Manually done the same task would take me half a few hours in the best case and a day in the worst case.

That's indeed a good example.

However I find their claim "I've lead teams of really competent engineers and I can leave them without supervision for months and come back and not feel like throwing away the entire code base." dubious. We all know how much effort it is to keep the quality of even small patches consistent.

Design, architecture, style and refactoring still require significant involvement. Providing only a description and a criteria will likely produce hopelessly messy code, which is also what you get with most corporate dev teams.

That's fair. In all honesty I'm already feeling challenged but given how much time I save I can set aside some time to keep myself sharp. I can learn more languages. Additionally, as pointed out by others, I'm trading coding effort for design and and strategy, which generally control business outcomes a lot more.

Having said that, I won't use AI for production system if I don't understand the programming constructs in enough detail.

I'm in the "haven't written any code in a while" boat ATM. I'd love to see examples of issues that are so big that they warrant reverting to manual coding.

My main issue has been the inconsistent quality across between model releases and the tendency to insert older APIs or documentation, especially with command line tools.

I can understand if the model struggles with a million line monolithic codebase with a decade of cruft but can't think of why it'd be too much of a pain with new codebases.

If you look at Australian IT companies they're management and consultant heavy. Roles like architects, review boards, program managers etc., exceed actual engineering roles. In such a set up it takes forever to get any real work done.

Then Australian real wages have also declined. So there is already low motivation to work beyond the minimum expectation.

Australia also has strong social security, at least until now, and there are plenty of odd jobs due to the real estate and tourism industries, so there is little pressure to survive unless you owe too much credit.

Automation and technology adoption also lags behind. For example, people still wait thirty minutes on a call to get an appointment with a doctor instead of making an online booking.

For all of these reasons productivity has been low and declining. An extra day off work is a strong stimulus for squeezing what gets done into the rest of the days.

In my opinion this is an extreme, just as the work yourself to death culture in India and China is. On a scale of a few decades economies where people worked agressively harder have grown tremendously but at the expense of the long-term wellbeing of the people. Places like Australia that were well off and felt little pressure to compete have sustained good quality of life but at the expense of having no competetitve edge anymore.

> Kiselev’s child reader is being treated as a participant in mathematics, not as a recipient of facts.

Not sure how the other makes this claim when the passage he himself cites is just another clever proof in the list of clever things that maths books throw at you:

> It is easy to convince oneself that there exist infinitely many prime numbers. Indeed, suppose the contrary, that the number of primes is finite. Then there must exist a greatest prime; let it be a. To refute this assumption, imagine the new number N formed by the rule N = (2·3·5·7···a) + 1, that is, the product of all the primes up to a, plus one… The first term is divisible by every number in the list 2, 3, 5, …, a, while the second (the unit) is not divisible by any of them. Hence there is no greatest prime, and so the sequence of primes is infinite.