Thanks for the writeup! It's fascinating that you can hook yourself up to a PC and get music played directly into your implant. I wonder what would happen if you shift all frequencies in a song (or PC microphone capture) into a much smaller range, proportinally? And then test some target ranges. Or you could slow down a song but keep all the frequencies?

AI can write such programs quickly.

Mi-Rae : this is exactly what I was thinking also.

Metaphor: If there is a problem with the lunar lander, or the Mars robots we can't easily go fix them, we can't send someone to "turn it off and back on again", so we find a work around instead. In your case we may not be able to change the physical connections easily, or heal the damage there, but we may still have some options. ...

Most of us here at hackernews are programmers or technologists, not medical experts. So if there is a medical reason / short, malfunction, or other problem with the underlying connection to your brain (or to your brain itself in that small connection area around the implant), that is outside of our expertise.

## Suggestion: This suggestion to "translate"/transform the speaking ranges DOWN into the ranges that you can actually hear is a great one. So you'd be looking/asking claude for software that would take any/all microphone input at normal speaking frequencies, and in real time, shift/adjust the frequency Downwards until the average speaking midpoint was below 70Hz . then if your brain can hear differences in the lower frequencies, it would just be a matter of training/learning them in those lower frequencies. It may still take some time to hear specific speech, but you need to get it into a range that you can actually detect some changes.

An audio recording or mixing technician may be able to help with suggestions about which frequencies usually contain male, female, child voices and things like "s" sounds or sharper hisses at the end of a sound which tend to be a higher frequency than the main words. But the main goal would be to make ANY speech detectable to your ears.

There may be multiple ways you could try stretching out or reshaping the word sounds

For faster testing, start with a single recorded word at normal ranges. As a starting place, Have a friend record a single word and verify that they can hear it on your computer when it plays (this is important!). Then you would try variations for your software to shift the word into lower and lower frequencies until it becomes legible/hearable to your brain. Once you are able to hear something then record the range that its shifting down into as the "sweet spot". Try a few alternatives above and below that spot to try to find your " best" hearing frequencies. Once you've done that, you'll just need to play that word over and over again until you are able to hear the edges of it, the articulation of the syllables which should sound like hills and valleys. In most languages including English and Korean, The beginnings and the ending sounds (and how abruptly they stop or start), like edges to a sword (wielded by an expert swordsman), or like the movements of a dance (sometimes a slow smooth movement, sometimes an abrupt stop, then movement again). Thise are the same kinds of things we do with our voices to make sounds. A shape of the syllable like a body dancing shape, the lights come up on the stage, then movement(air&vibration) starts in a certain way, then movement (fast or slow) into another mouth shape. Usually the beginning and middle are particularly important, many dialects and creoles tend to drop or slur the ends of words, since if you have already heard the beginning and middle of the word so many times, then you already know the end part. Similarly, if you watch your favorite dance or ballet over and over again, you remember and can know the shape of the body in the next beat before the dancer gets there.

Best of luck on this, please report back if you make any progress, we are all cheering for you, and we hope you're able to make progress!

Reminder: to retrain your brain on different inputs takes time. Make sure you can hear something in the changes, but then when you're in the "sweet spot" range, give it time. Listen hard! Be a learner and always be expecting for that dancers shape to come through. Spend time doing it every day, maybe with a blindfold so that you can focus ONLY on the sounds, otherwise your smart brain will try to work around it, picking cues from visuals.

I've never thought of trying that approach before. It's definitely worth giving a try on my day off. Thank you for the great idea.

I wonder if over time the two will simply merge and become indistinguishable from one another. The human writing is already changing, as the author has observed. The LLMs will follow the change, and soon meet us in the middle!

Or maybe the humans will do the opposite and shift to Sardukar-style communication instead? LLMs can easily match that as well.

That style I would very much prefer to all the fluff people write today and were writing even before AI.

The other day I was writing a blog post summarizing how I use an adaptation of a context handling technique and I wrote the following three lines:

* Every user run becomes a memo

* Days summarize, rolled to weeks, rolled to months - all written once

* Today's runs are shortened and placed into running context past the cache break

Then while reading it before publishing it, I thought it sounded super LLM-y. I didn't rewrite it because I jammed that stuff together before hitting publish because I was in a hurry but now that I go re-read it, it's somehow worse. Dear god. It's happening.

The classic thing is this business with inanimate objects verbing metaphors or becoming other objects. It's a strangely uncanny valley, but I suppose I read so much of this with Claude that it gets stuck in my head!

> Indeed, the space of mathematical problems remains infinite;

> and the difficulty level of such problems is unbounded.

Are these two sentences true? Is there a way to prove or disprove them?

Could the space of mathematical problems be finite and their difficulty level bounded?

I think that vertical is actually a step function that is still limited by the human bottleneck of deciding what to build and making sure that it actually gets built. Give the AI an underspecified task, you get what you asked for and also what you didn't ask for, in good old Genie fashion.

Hill-climbing problems seem to escape this limitation. But I wonder what will happen when the AIs start acing benchmarks faster than you can write new benchmarks, and when you run out of hard problems to solve because you rely on humans formulating them.

The real vertical will come when the AI gets access to physical world instead of bash; and when the human is finally out of the loop.

Given what you write here, would you feel safe giving AI agents access to the physical world? Which kind of safeguards would you implement?

The VC-backed pressure will ensure that all kinds of safeguards will be skipped, regardless of my feelings! Solving millenium math problems in a few hours might piss off some mathematicians. But an AI that's autonomously looking for cold fusion or a cure against cancers might cause planetary-scale desaster on the way there.

The only reasonable safety measure is to always keep humans in the loop, and to give them time to understand and consolidate all advances, to actually go slow.

the thing to consider as well is when tooling catches up the 20m that still get it will accelerate

I think unsafety, distillation, and China catching up in chipmaking are inevitable.

Just like how it was inevitable for SoTA LLMs to ignore copyright.

The actual challenge isn't how to prevent all these, but how stay on top.

And to stay on top it is inevitable to train unrestricted models. Anthropic is fighting windmills.