brandoncarl

With purely quadratic attention, the cost of the token at 372K is 87% more than the token at 272K.

There are attention mechanisms that help to mitigate this, but you can clearly see in the chart that Tibo posted that the attention mechanisms they are using are still scaling quadratically.

This becomes an optimization problem for token t:

min f[cost, quality loss] = compaction cost[t] + compaction quality loss[t] + token cost[t]

HN -

The topic of LLM jailbreaking is an important one. As machines increase in power their ability to benefit society increases, as does their ability to harm society.

This paper intends to address the question: "Is it possible to create a perfect universal guardian against LLM jailbreaks?". This means:

Universal: applies to all prompts and input combinations Perfect: is able to successfully distinguish harmful versus not harmful Computable: is feasible to compute We show that this combination does not exist and Request for Comment.

A more feasible approach involves post-hoc filtering of results, which involves an easier problem of decideability. In order to implement such approaches, both reasoning traces an intermediate code changes would likely need to be hidden from the user until the results are determined.

Note: this paper was written in conjunction with ChatGPT 5.5 Pro. The topic of the reducibility of such problems to The Halting Problem has been top of mind for me for some time. In one sense, it is a trivial conclusion. Yet it is one of high importance.

Markdown has emerged as the "lingua franca" of LLMs. It is structured enough to convey meaning, and yet readable everywhere.

However, as document lengths increase: 1. It is wasteful to put the entire context in memory 2. It can reduce accuracy by providing less relevant information 3. Increase the possibility of errors during edits

Inspired by Cloudflare's Markdown Content Type and Code Mode, we propose a lightweight document object model for Markdown to: 1. Reduce costs 2. Increase the accuracy of locating key sections 3. Increase speed 4. Increase the accuracy of editing documents

https://github.com/brandoncarl/markdown-dom

Thank you

I used this for the camera angle. Works really well.

https://www.amazon.com/gp/product/B07F1DB2F7/ref=ppx_yo_dt_b...

On timing – I just had to start planning a bit better. In worst case I'd start the meeting without video and turn on a minute or two in. Well worth it for me.

I use Camo (lifetime $79 license) and an iPhone SE I picked up on eBay for $200.

I've had dozens of people comment how clear my camera is. Many now purchased a similar setup. I used OBS for a long time since it was free, but the quality wasn't close.

If you're in a line of work that requires you to stream it's a worthwhile investment by a great company that just keeps getting better.

This is the closest. https://assets.ctfassets.net/vyse88cgwfbl/1np5dpcwuHrWJ4AgUg...

As you can see, the rates don't map to any significant overcollateralization.

You're absolutely right regarding from market highs.

To do this correct you'd need purchases and sales versus assets.

All that said, it's why I put a rough estimate of 30% (instead of 60%) and hadn't included any losses in broader commercial paper (which has dropped). So the overall estimate is defensible.

Also agree on Basel 2-like risk. The primary issue with Tether is the marketing around stability and the transparency. If it was treated like any other "money market" with better NAV reporting, position reporting and rules around withdrawals then we'd mitigate some of the risks.

Important to note that a currency peg is different than a currency. The latter free floats, the former does not.

Investments were this way in the 1800s and early 1900s.

A lot of protections have been put into place since then. There's a reason we prosecute Ponzi schemes, offer FDIC and SIPC insurance, stress test banks, etc.

I follow your theory, but as of the last audit that wasn't close to true.

Tether's underlying assets are almost certainly worth less than $1. You can see this by taking each asset class and marking it to market.

I've provided details here: https://twitter.com/brandonjcarl/status/1536310742261047296?...

By itself that's not enough for Tether to "fail". It does mean that the asset is in a position of very high risk and subject to a run.

It's fantastic to see great research combined with great tutorials – bravo. You all have bitten off an incredibly hard problem with pragmatism and tenacity. Keep up the great work.

To the authors: did you any of your own recordings? I've used my own and clips online, in WAV and other formats, at various sampling rates.

All of the results come back gibberish. The results in the training data seem just fine. Curious if you've tested the above to ensure it didn't overfit.

Earning per share....

As I understand they do have a working prototype. I believe their chip is 3mm and works in pockets, etc. There's been so much hype, so little delivered in this area that I'm finding it hard to sift the wheat from the chaff.

Via the Lodash CLI you can "lodash underscore". This results in 6.7kb gzipped. In my experience, jdalton is a very solid coder and the lodash package is excellently maintained.

If anybody has interest in working on some of the Econ-related issues, please tweet me at @brandonjcarl

How can you possibly say "no offense" and then proceed to make fun of somebody? Simply using that phrase doesn't make it any more ok.

Obviously hard to tell, but you can extrapolate a bit.

At $100mm in revenue, and an assumed $60 order price, you've got 1.6mm orders. Assuming 25% of people pay for faster delivery, you've got gross revenue per order of $4.50, or gross revenue of $7.2mm. If Instavart takes a 30% share (similar-ish to Uber and iTunes), you're looking at $2.2mmin revenue after paying contractors.

So that would leave you with 2.2% of sales as your margin after contractors. This obviously changes if they mark up the groceries or if the assertions are different.

Out of interest, a 15 P/E on 2bb would correspond to a 60x grow in their market (ish). They currently represent about 100mm of Whole Foods 15bb take, or about 0.67%. A 60x increase would take them to managing 40% of Whole Foods orders, given no expansion into other grocery chains.

This is a lot of hand-waving, so please take it with a grain of salt.

Yes, Tim Cook's an operational genius. However those cell phone plans are starting to change, which I think makes things a bit more complex for Apple.

I'm surprised that nobody seems to reference Dropbox or Github. Both companies are well-respected, successful, good at attracting talent, and choose to use CoffeeScript. I would wager that both generally manage programming projects that are larger than the majority of the HN community does.

Aventura Labs - New York, NY

Aventura Labs is an early stage company whose mission is to assemble great algorithmists, programmers, and designers to solve hard problems.

I've spent the past year bootstrapping the first product, Just the List, which focuses on making wedding planning fast and fun. The product is near complete, but also has an awesome roadmap of upcoming features. Beta-tester feedback has been outstanding. For more info, please check out the soon-to-be-released home page can be found here:

http://jtlist.co/1fE7MWB

You'll be joining on as number two at the company. You should love learning, solving hard problems, and be interested in culture building. You should believe that unicorns are made, not born :) You should also be familiar with most of the technologies in our stack: Node/Express, Javascript/Coffeescript, MongoDB, Redis and Backbone.

I'm hiring a variety of roles which may include summer interns, developers, a CTO, and a community manager. We'll be working on a pretty full combination of improving/writing new algorithms, performing data analysis, automating front-end testing (backend testing is largely completed), and building intuitive and beautiful user experiences.

I've open-sourced the underlying task queue at https://github.com/brandoncarl/taskco. I'd like to open-source a few of the other technologies as the team grows.

Please contact me at brandon.carl@justthelist.com for more details.

You've got to love that! Great entry point.

Does anti-trust ever become an issue in that case? What about availability of cash (they just spent 35% of theirs)? What about properly valuing users?

While I'm following you ideologically, I'm not sure that this idea bears fruit in evidence. The computer science community still uses large numbers of algorithms (via libraries) that were written by academics, in peer-reviewed journals, in Fortran, in the 1970s.

Even currently, a tremendous amount of the cutting-edge work in optimization and others is done at the university level: METIS for graph partitioning, and convex optimization work being done by Stephen Boyd et al. They care deeply about the implementation because quality and speed of solution are of utmost importance. A number (like Geoffrey Hinton and Sebastian Thrun) have moved into the private sector, but after contributing a lot via academia.

That's fair, but keep in mind that the CDC this year revised their estimates of people with Lyme 10x. This suggests that they've had a strained understanding of the problem.

http://www.pbs.org/newshour/bb/health/july-dec13/lyme_08-20....

Would you agree that a medical journal's impact factor is more important for new research than a summary? I would think that for a summary the quantity and quality of the references is a bigger factor.

As far as the clinicians, having a lot of family experience with the disease, I've seen things far differently. This in addition to support groups, phone calls, etc. We're not going to get anywhere debating experience, just letting you know that it happens a lot.

I've watched my mother suffer from this for 20+ years. She initially had a doctor that treated her for 3 years with very strong antibiotics. She made remarkable progress. The doctor left practicing for personal reasons and subsequently she couldn't find good medical help. Her condition deteriorated over the years.

While the CDC doesn't seem to dismiss symptoms, I've watched dozens of doctors dismiss her. I've received calls from dozens of people who all have the same set of symptoms. I've also seen people get better only after years of antibiotic treatment, and slowly at that. My friends in medical school have indicated that they receive about two hours of Lyme-related training during their education.

The issue that you have is three-fold:

1. Ticks often carry multiple infections (Borrelia Burgdoferi, Bartonella, Mycoplasma, and Babesiosis). Each of these infections can leave the immune system in a dysfunctional state. Most studies only focus on treatment of BB, leaving the other infections untouched.

2. Given that, for the toughest cases, years of antibiotics may be needed, it is extremely difficult to keep studies "controlled". i.e. a week long study may not have much variability, but over a two-year study, a lot happens. The noise increases relative to the signal.

3. BB has very high genetic variability. This leads to the potential for larger numbers of strains, with over 200 thought to be in existence.

I'm not contesting that this isn't well-understood. In fact, I fully agree with that. But statements like there is little to no evidence that Chronic Lyme exists are inappropriate. This is an area that has historically been underfunded, and it has a pretty rich history of controversy. We can fight over nomenclature all day long, but that doesn't change the fact that a lot of people are suffering, and the broader medical community is woefully uninformed.