HNHacker News
TopNewBestAskShowJobs

ljosifov

660 karma · joined June 6, 2020

Now - ML/AI, llms, agents, meta-learning, auto-everything, forecasting, research & engineering, sciences. Previous - systematic trading, quant trading research & development. Previous^2 - speech recognition in noise, speech synthesis, machine learning. In Harpenden UK, from Skopje MK.
submissionscomments
ljosifov··on Updates to Consumer Terms and Privacy Policy
Not at all, on the contrary, I chose my words carefully ("mildly sociopathic OF ME") as to avoid casting shade on others. Saying "this is how I feel", so to preclude judging others. Everyone makes their own choices, and that's fine.

Boundaries - yes sure they exist. I don't have my photo albums open to the world. I don't share info about family and friends - I know people by default don't want to share information about them, and I try to respect that. Don't share anything on Facebook, where plenty share, for example.

At the same time, I find obstacles to data sharing codified in the UK law frustrating. With the UK NHS. 1) Can't email my GP to pass information back-and-forth - GP withholds their email contact; 2) MRI scan private hospital makes me jump 10 hops before sharing my data with me; 3) Blood tests scheduling can't tell me back that schedule for a date failed, apparently it's too much for them to have my email address on record; 4) Can't volunteer my data to benefit R&D in NHS. ("here are - my lab works reports, 100 GB of my DNA paid for by myself, my medical histories - take them all in, use them as you please...") In all cases vague mutterings of "data protection... GDPR..." have been relayed back as "reasons". I take it's mostly B/S. They could work around if they wanted to. But there is a kernel of truth - it's easier for them to not try share, so it's used as a cover leaf. (in the worst case - an alibi for laziness.)

I'm for having power to share, or not share, what I want. With Google - I do want them to know about myself and use that for my (and theirs) benefit. With the UK gov (trying to break encryption) - I don't want them to be able to read my WhatsApp-s. I detest UK gov for effectively forcing me (by forcing the online pharmacy) to take a photos of myself (face, figure) in order to buy online Wegovy earlier today.

ljosifov··on Updates to Consumer Terms and Privacy Policy
If they leaked bank accounts numbers, or private keys - I would be worried. That has not happened in the past.

About myself personally - my Name Surname is googleable, I'm on the open electoral register, so my address is not a secret, my company information is also open in the companies register, I have a a personal website I have put up willingly and share information about myself there. Training models on my data doesn't seem riskier than that.

Yeah, I know I'd be safer if I was completely dark, opaque to the world. I like the openness though. I also think my life has been enriched in infinitely many ways by people sharing parts of their lives via their data with me. So it would be mildly sociopathic of me, if I didn't do similar back to the world, to some extent.

ljosifov··on Updates to Consumer Terms and Privacy Policy
Excellent. What were they waiting for up to now?? I thought they already trained on my data. I assume they train, even hope that they train, even when they say they don't. People that want to be data privacy maximalists - fine, don't use their data. But there are people out there (myself) that are on the opposite end of the spectrum, and we are mostly ignored by the companies. Companies just assume people only ever want to deny them their data.

It annoys me greatly, that I have no tick box on Google to tell them "go and adapt models I use on my Gmail, Photos, Maps etc." I don't want Google to ever be mistaken where I live - I have told them 100 times already.

This idea that "no one wants to share their data" is just assumed, and permeates everything. Like soft-ball interviews that a popular science communicator did with DeepMind folks working in medicine: every question was prefixed by litany of caveats that were all about 1) assumed aversion of people to sharing their data 2) horrors and disasters that are to befall us should we share the data. I have not suffered any horrors. I'm not aware of any major disasters. I'm aware of major advances in medicine in my lifetime. Ultimately the process does involve controlled data collection and experimentation. Looks a good deal to me tbh. I go out of my way to tick all the NHS boxes too, to "use my data as you see fit". It's an uphill struggle. The defaults are always "deny everything". Tick boxes never go away, there is no master checkbox "use any and all of my data and never ask me again" to tick.

ljosifov··on If you have a Claude account, they're going to train on your data moving forward
Excellent. What were they waiting for up to now?? I thought they already trained on my data. I assume they train, even hope that they train, even when they say they don't. People that want to be data privacy maximalists - fine, don't use their data. But there are people out there (myself) that are on the opposite end of the spectrum, and we are mostly ignored by the companies. Companies just assume people only ever want to deny them their data.

It annoys me greatly, that I have no tick box on Google to tell them "go and adapt models I use on my Gmail, Photos, Maps etc." I don't want Google to ever be mistaken where I live - I have told them 100 times already.

This idea that "no one wants to share their data" is just assumed, and permeates everything. Like soft-ball interviews that a popular science communicator did with DeepMind folks working in medicine: every question was prefixed by litany of caveats that were all about 1) assumed aversion of people to sharing their data 2) horrors and disasters that are to befall us should we share the data. I have not suffered any horrors. I'm not aware of any major disasters. I'm aware of major advances in medicine in my lifetime. Ultimately the process does involve controlled data collection and experimentation. Looks a good deal to me tbh. I go out of my way to tick all the NHS boxes too, to "use my data as you see fit". It's an uphill struggle. The defaults are always "deny everything". Tick boxes never go away, there is no master checkbox "use any and all of my data and never ask me again" to tick.

ljosifov··on Ask HN: Why hasn't x86 caught up with Apple M series?
On selling of the ARM stake - asked ChatGPT:

Q> And latter - was it that it looked that Apple maybe in danger of going under, and then they sold their ARM stake and got a cash injection that way?

A> And yes. In the late-1990s turnaround, Apple sold down its ARM stake in multiple tranches after ARM’s 1998 IPO, realizing hundreds of millions of dollars that helped shore up finances (alongside the well-known $150 million Microsoft deal in Aug 1997).

ljosifov··on Ask HN: Why hasn't x86 caught up with Apple M series?
Were they ARM investors at the time they needed CPU for Newton? Was that before or after e.g. iPaq PDA-s? And latter - was it that it looked that Apple maybe in danger of going under, and then they sold their ARM stake and got a cash injection that way?

I remember iPaq PDA fondly. Wrote a demo to select a song from a playlist with few thousand author-album-song with voice query. The WiFi add-on was a big plastic "sleeve", that the iPaq slid into, not the other way around. Could run the ASR engine for about whole 10 mins before it drained the battery flat, haha. :-)

ljosifov··on Ask HN: Why hasn't x86 caught up with Apple M series?
Plenty of excellent comments about the companies - e.g. Apple vertical closed mobile 1st, while Microsoft horizontal open desktop 1st; decades of work by many thousands of people went into optimising many tiny advantages, aka tricks - but can't help but think back of pre-history. Where Intel was always more-is-more, while ARM was always less-is-more. Intel was winning for the longest time. Never expected to see non-x86 competitive single core integer performance tbh. And in the pre-pre-history, one generation further back, tiny 6502 1MHz and mostly totally 8 bit only, could about keep up with Z80 4MHz and his almost-aspiring-to 16 bit registers. Always made me wonder somewhat - "whut, how come??"
ljosifov··on Wikipedia loses challenge against Online Safety Act
US should slap travel bans on UK politicians travelling to Disney parks and similar in Florida with their families. And/or with their older children visiting NYC. The combined pressure of the wives and their children, will knock sense in their thick skulls quickly. In the sense of - being stupid is not cost free. Atm it's cost free for them, and costly for me.
ljosifov··on I tried every todo app and ended up with a .txt file
Good man. Everyone eventually reaches the same year zero: a text file.

Then adds structure back, as it suits their persona. Not too much, not too little, just right - goldilocks. It's very personal, even more than a smartphone.

For me - $ githome add logBook to git $HOME solves versioning and replication:

giho() { (cd "$HOME" && git --git-dir="$HOME"/.githome/ --work-tree="$HOME" "$@";) } # prior must $ git init --bare $HOME/.githome

And in the logBook structure currently at:

1. Sections FIXME, TODO, DONE, DONTDO. Keep them vi searchable /^SECTION$.

2. Entry start searchable /^-(space).

3. Entry end separator from next is empty line searchable /^$.

4. New items add at the top, push old items down.

5. Items move wholesale, no change on between sections move.

6. Items spending too long in TODO moved into DONTDO.

7. No new items added in TODO if FIXME is not empty.

8. If really really need to add to TODO - then move blocking FIXME entries to TODO.

9. Above are rules of thumb - break them with a reason, don't break them without reasons.

ljosifov··on I want everything local – Building my offline AI workspace
In the same boat. I love running things localhost. It's been great fun, and I learned tons I didn't know before. I know remote models API-s are a must for any serious work where tons is to be done, produced fr. Still it warms my heart every time llama-server runs on, and serves my aging mbp. Recent MoEs run great on macs with loads of v/ram, and the power efficiency is scarcely believable.
ljosifov··on If you're remote, ramble
I stuff those in my logBook nowadays - a single ascii text file. Started as you say with notes on random scraps of paper. To relieve my mind of carrying that burden, when I have something more pressing to do. But yeah these half-thoughts, intuits etc have showed useful over time? Some made it into TODO, and latter even into the DONE entries. Even if mostly their final destination is DONTDO. :-)
ljosifov··on Writing is thinking
It's possible to not use Google for search. I switched to Perplexity many months ago. Almost never come back to Google. No one is forcing me to use a worse option, when a better one exists. Just use something else - easy.
ljosifov··on LLM Inevitabilism
I don't think it's inevitable, for very few things are really inevitable. However, I find LLM-s good and useful. First the chat bots, now the coding agents. Looks to me medical consultation, 2nd opinion and the like - are not far behind. Enough people already use them for that. I give my lab tests results to ChatGPT. Tbh can't fault the author for motivated reasoning. Looks to me it goes like: this is not a future I want -> therefore it should not happen -> therefore it will not happen. Because by the same motivated reasoning: for me it is the future I want. To be able to interact with a computer via language, speech and more. For the computer to be smart, instead of dumb, as it is now. If I can have the computer enhance my smarts, my information processing power, my memory - the way writing allows me to off-load from my head onto paper, a calculator allows me to manipulate numbers, and computer toils for days instead myself - then I will probably want for the AI to complement, enhance me too.
ljosifov··on Bret Victor on why current trend of AIs is at odds with his work
I get what you are talking about. My gripe with that is - yeah, would be indeed great if we could at some point get the structure to such deep level, as to write down pen on paper on one page, a string of maths symbols that is a good enough description. However - it's possible that for many things, that's not possible. Suspect maybe not possible in e.g. biology. Possible that the great success of physics of the 20-th century over-indulged us. So our expectations are out of kilter with realities of our world.

Fwiw I personally describe than as white, not black boxes. For we know, and can trace back every single bit of the output, back to the input. That does not help us as much as we'd like though. When drilling down into "why did the model answer wrongly 1, and not rightly 2", it comes down to "well, it added one trillion small numbers, and the sum came close to 1, but didn't reach 2". Which is unsatisfactory, and your "understanding" v.s. "comprehension" delineates that nicely.

Maybe more productive to think of them more "artefacts", less "mechanical contraptions". We shape them in many ways, but we are not in complete control of their making. We don't make them explicitly with out hands: we make a maker algorithm, and that algorithm then makes them. Or even "biological", grown artefacts. Given we don't control the end result fully. Yes we know and apply the algorithm that builds them, but we don't know the end result before hand, the final set of weights. Unlike say when we are making a coffee machine - we know all the parts to a millimetre in advance, have it all worked out pre-planned, before embarking on the making of the machine.

ljosifov··on Geometry from Quantum Temporal Correlations
excellent playlists - thanks!
ljosifov··on Geometry from Quantum Temporal Correlations
A recent Vedral (one of the authors) talk -

Decoding quantum reality - with Vlatko Vedral @ The Royal Institution (4-Mar-2025; 59:26)

https://youtu.be/70FhS6NAbuA

(I mostly watch while reading the running transcript these days - https://www.appblit.com/scribe?v=70FhS6NAbuA)

ljosifov··on Beware of Fast-Math
I can imagine sth like: if (bid ask blah blah) { send order to buy 10 million of AAPL; }
ljosifov··on Deepseek R1-0528
It will recombine the existing resources in new ways. Neanderthals had access to exactly the same natural resources as we have now. Obviously we do much more with what we both got, then they ever did. Obviously it's not only the availability of some atoms or molecules, but what one does with them, how one recombines them in novel ways. For that one needs knowledge and energy. And the later mostly turns out can be derived from the the former too.
ljosifov··on Deepseek R1-0528
I'm assuming the Intelligence Revolution, multiplying Human Intelligence with machines, will have the same effect as the Industrial Revolution had, on multiplying human physical strength. That multiplied the GDP by a factor of ~20 times, hockey stick like, in a fairy short time, a century or two.
ljosifov··on Deepseek R1-0528
You may be right about the damage (will not dispute it even if I personally doubt it) - but what about the amount of good that it can do too? When deciding "what is to be done now" under uncertainty, we typically look at both sides of the ledger, the upsides in addition to the downsides.

Assume for a moment, that the current AI is teaching us that compute transforming data → information → knowledge → intelligence → agency → ... → AGI → ASI, is all there is to Intelligence-on-Tap? And imagine an AI path opens to AGI now and ASI later, where previously we didn't see any. Seems a bad deal to me, to frustrate, slow down, or even forego the 2050-s Intelligence Revolution that may multiply total human wealth by a factor of 10 to 20 in value, the way the Industrial Revolution did in the 1800-s. And we are to forego this, for what - so that we provide UBI to Disney shareholders? Every one of us is richer, better off now, than any king of old. Not too long ago, even the most powerful person in the lands could not prevent their 17 miscarriages/stillbirths/child_deaths failing to produce an heir to ascend the throne (a top priority that was, for sure for a king+queen). So in our imagined utopia, even the Disney shareholders are better off than they would be otherwise.

ljosifov··on Deepseek R1-0528
:-) I like the symmetry of this. If I want to keep my creations outside the hands of others, I can keep them private. I don’t have to publish these words or broadcast them to the world. I could write this on my laptop, save it in a file, and keep it to myself. Fine.

However, once these words are broadcast—once they’re read, and the ideas expressed here enter someone else’s mind—I believe it’s only fair that the person on the receiving end has the right to use, replicate, or create something from them. After all, they lent me their brain—ideas that originated in my mind now live in theirs.

This uses up their mental "meat space," their blood sugar, and their oxygen—resources they provide. So, they have rights too: the right to do as they please with those ideas, including creating any and all data derived from them. Denying them that right feels churlish, as if it isn’t the most natural thing in the world.

(Before people jump on me:- Yes, creators need to be compensated—they deserve to make a living from their work. But this doesn’t extend to their grandchildren. Copyright laws should incentivize creation, not provide luxury for the descendants of the original creator a century later.)

ljosifov··on Why does the U.S. always run a trade deficit?
Haha :-) - fine, hope no offence taken. :-) (none was intended) It came with the flow, flow was - US-... -Government, -Citizens, -Dollar, and then when it came to "people of the world that are not US citizens" (but are users of US dollars) programatically is seems to my compute mind, it came as "non-US-ians" fits that set of people well, and plus tallies with the "US-this-and-that" of the wordings prior. I'm originally from the Balkans, and so cognisant and got some experience with "verbal mines" and nuances in all manner of names and naming-s. I intuit what you are saying. :-) thanks
ljosifov··on Gemma 3n preview: Mobile-first AI
On Hugging face I see 4B and 2B versions now -

https://huggingface.co/collections/google/gemma-3n-preview-6...

Gemma 3n Preview

google/gemma-3n-E4B-it-litert-preview

google/gemma-3n-E2B-it-litert-preview

Interesting, hope it comes on LMStudio as MLX or GGUF. Sparse and or MoE models make a difference when running on localhost. MoE Qwen3-30B-A3B most recent game changer for me. Activating only 3b weights on the gpu cores of sparse Qwen3-30B-A3B, rather than comparable ~30b of dense models (Qwen3-32B, Gemma3-27b, GLM-{4,Z1}-32B, older QwQ-32B), is a huge speedup for me: MoE A3B achieves 20-60 tps on my oldish M2 in LMStudio, versus only 4-5 tps for the dense models.

Looking forward to trying gemma-3n. Kudos to Google for open sourcing their Gemmas. Would not have predicted that the lab with "open" in the name has yet to release even v1 (atm at 0; disregarding gpt-2), while other labs, more commercial labs, are are at versions 3, 4 etc already.

ljosifov··on Why Does the U.S. Always Run a Trade Deficit?
Because there are 1) willing exporters of US dollars USD (the US federal government USG ultimately, and US citizens in the final analysis); and 2) willing users of USD - us non-US-ian people around the world.

When I sell stuff to someone in the UK, then pounds are fine as means of exchange. When I sell the same someone around the world, then the transaction not being domestic, the cash part that is 1/2 of it (or 1/4 if you want to look at it in a way of transaction = quartet of {stuff,cash,buyer,seller}) and it can be in any currency of any country. It helps if it is the country of the buyer or the seller, but in general it is neither.

I prefer US Dollar to Zongoan Vonga, b/c:

1) USG typically does not devalue their USD much, I don't have to rush to exchange the USD for something else. Or if you want - most others devalue theirs more.

2) USG has not deleted and re-issued their USD even once so far afaik. Other countries do it more often.

3) Everyone else uses USD already. It's easier if it's one currency sufficing for all trade.

4) Often I don't produce oil, and oil is easiest to pay by with USD. Not impossible with others, just more difficult.

5) US banks allow me to hold USD in accounts outside of US. They don't ban me to take the USD in- or out- of US.

6) With USD in hand I can buy lots of stuff US produces, plus I can buy US assets - property, financial assets etc. US does not discriminate against me johnny-the-foreigner much, they are mostly relaxed in that respect. Other countries have all manner of rules that amount to: they'll basically steal their currency off you, or make it worthless to you by making you unable to claim the goods & services that are nominally sold in that currency.

Before people pipe in how US is terrible in this and that - above is all *relative* to all other options all other currencies available. This is from to top of my head, I'm sure I made some error and there is more.

ljosifov··on Run LLMs on Apple Neural Engine (ANE)
This more than anything feels emblematic to me, that Apple executives are brain dead when it comes to software. AI seemingly being a step too far. (in the s/w direction.) While they could at some level grok classical s/w, NN-s are that Terra Incognita where Apple executives can not possibly follow in. It's just too strange and mysterious world for them to be able to effectively decide or execute on anything. Worked (20yrs ago) in an industrial R&D lab of a mostly h/w manufacturer for 3 years. It looked to me that the worlds of h/w and s/w, the mindsets, diverged pretty quickly on all important considerations of "what should happen next".
ljosifov··on Qwen3: Think deeper, act faster
I'm enjoying this ngl. :-) Alibaba_Qwen did themselves proud--top marks!

Qwen3-30B-A3B, a MoE 30B - with only 3B active at any one time I presume? - 4bit MLX in lmstudio, with speculative decoding via Qwen3-0.6B 8bit MLX, on an oldish M2 mbp first try delivered 24 tps(!!) -

24.29 tok/sec • 1953 tokens • 3.25s to first token • Stop reason: EOS Token Found • Accepted 1092/1953 draft tokens (55.9%)

Thank you to LMStudio, MLX and huggingface too. :-) After decades of not finding enough reasons for an MBP, suddenly ASI was it. And it's delivered beyond any expectations I had, already.

Did I mention I seem to have become NN PDP enthusiast, an AI maximalist? ;-) I thought them people over-excitable, if benevolent. Then the thought of trusting Trump-Putin on decisions like thermo-nuclear war ending us all, over ChatGPT and its reasoning offspring, converted me. AI is our only chance at existential salvation--ignore doom risk.

ljosifov··on Qwen3: Think deeper, act faster
Amusingly enough, people writing stuff like the above, to my mind come over as doing what they are accusing LLMs of doing. :-)

And in discussions "is it or isn't it, AI smarter than HI already", reminds me to "remember how 'smart' an average HI is, then remember half are to the left of that center". :-O

ljosifov··on The US stops sharing air quality data from embassies worldwide
I don't necessarily disagree on the need - but it's not my (re)definition.

Having said that - I rather like the use. "Stupid" is part of quartet. All humans are categorised in four categories: intelligent (I), helpless (H), stupid (S), bandit (B). In X-Y coordinate system, by quadrant Q, actions taken by Tom, that affect both Tom (on X-axis) and Dick (on Y-axis):

  - Q1 (+,+) Tom gain and Dick gain => Tom was intelligent (I).

  - Q2 (-,+) Tom loss and Dick gain => Tom was helpless (H).

  - Q3 (-,-) Tom loss and Dick loss => Tom was stupid (S).

  - Q4 (+,-) Tom gain and Dick loss => Tom was a bandit (B).
Specifically on the S-Stupid in quadrant Q3 (-,-):

  + There is no upper bound on the amount of stupidity that can exist within any particular individual.

  + We always underestimate the number of stupid people.

  + The probability of a person being stupid is independent of any other characteristic of that person.

  + A stupid person is someone who causes damage to another person, or a group of people, without any advantage accruing to himself (or herself), or even with some resultant self-damage. (Golden law of stupidity)

  + Non-stupid people always underestimate the damaging power of stupid people, and that it's impossible to predict the actions of the stupid.

  + A stupid person S is the most dangerous person (compared to I, H, B), and crucially stupid S is more dangerous than a bandit B.
This is all on individual one unit level. Another relevant observation I can think of is Bismarck-ian most dangerous general — stupid and energetic.

Bonhoeffer is the one to make a connection to morality. That looks very important observation to me. And that it happens in groups, like a mind virus-I can see that. Sabine's video is all about collective stupidity. (Extraordinary Popular Delusions as a mirror image to Wisdom of crowds.)

ljosifov··on The US stops sharing air quality data from embassies worldwide
IDK whether you know it, but in case you don't—you articulated Cippola's definition "stupid is one causing harm to others while not deriving any gain, even taking a small personal loss" above. If you are interested further, I encourage you to check his pamphlet "The Basic Laws Of Human Stupidity". (is "book" too big word for it?)

I must say I got a lot out of it and it's followups. It became much easier for me to understand the world around me, what is happening. And even predict what's to happen sometimes.

What you observe with Trump admin: 1) it's stupidity; 2) it's not low IQ or lack of rationality; 3) it's lack of morals, not caring about good/bad, or even considering good what the rest consider bad (and vice versa); 4) it's social more than individual, it's a "mind virus". (ha!-Musk)

Bonhoeffer‘s "Theory of Stupidity" https://youtube.com/watch?v=ww47bR86wSc (bad morals not low IQ, mind virus social more than individual)

Sabine Hossenfelder "Collective Stupidity -- How Can We Avoid It?" https://youtube.com/watch?v=25kqobiv4ng (more cases, much more analytical, only about the group aspect not individual)

ljosifov··on Grok3 Launch [video]
People like getting scared. That's how they pay billions of $$$ every year to watch mostly cr*ppy horror movies.

GM has been a joke for years now. At some point his ramblings reached a GPT3.5 level, and have not improved since.

It's inditement on humans' logic and reasoning to give non-zero time to GM. Alas—we are human, we are both collectively clever (Wisdom of Crowds) and collectivelly stupid (Extraordinary Popular Delusions).

← PreviousPage 3 of 6Next →