HNHacker News
TopNewBestAskShowJobs

gojomo

30,216 karma · joined February 24, 2007

Gordon Mohr, maker of software.

Twitter: http://twitter.com/gojomo

Project: http://thunkpedia.org

Idea blog: http://memesteading.com

Older blog: http://gojomo.blogspot.com

Homepage: http://xavvy.com (username @ here for email contact)

My HN peeve is formulaic downbeat comments, like: "How is this news, I already knew this!" "…Betteridge's Law…" "I stopped reading at…"

submissionscomments
gojomo··on CDC ditches 5-day Covid isolation, argues Covid is becoming flu-like
Doesn't seem to have been as durably polarizing elsewhere, as far as I can tell.

EG: even European countries that were more draconian earlier in the pandemic returned to relative normalcy, with less fear/debate around school reopenings, mandatory testing, mandatory masking, etc, sooner than the US.

In the US, sticky "showy extreme COVID vigilance is my essential identity as a good citizen" or "showy nonchalance is my essential identity as a free person" partisanship lingers.

gojomo··on CDC ditches 5-day Covid isolation, argues Covid is becoming flu-like
Unless you specifically infect a treatment group in an RCT, your 'recently had COVID' and 'no recently documented COVID' groups will vary in ways you're unlikely to be able to fully control-for, no matter how hard you try.
gojomo··on Apollo calls AI a 'bubble' worse than even the dotcom era
But when the gold rush is reaching a newly-discovered lode, at the end, there's real new gold!

That they both eventually exhaust doesn't make them the same; one is fumes, the other involves a real expansion in the stock of usable resources.

People's cynicism here has them ignoring real enduring positive shocks in the economy – even though that's been the nonstop history of our tech industry. A few suppliers transient stock valuations may be, at times, a 'bubble'. But the fundamental (& irreversible advances) in AI technology aren't the bubble.

gojomo··on Third Base (ternary notation for computing) (2001)
If current top submission (https://news.ycombinator.com/item?id=39535800) turns out to be a meaningful advance, this old intro to ternary notation & its potential efficacy for computing may be of renewed interest.
gojomo··on The Era of 1-bit LLMs: ternary parameters for cost-effective computing
That's not a 'bit' ("Binary digIT"). It's closer to a 'trit' ("TeRnary-digIT"). Specifically, ternary digits spanning {-1, 0, 1} (rather than the usual {0, 1, 2} in a base-3 numbering system) are 'balanced ternary'.

A great intro to the theoretical reasons ternary might have some promise in computing is this 2001 article from 'American Scientist', "Third Base", which quotes Knuth calling balanced-ternary "perhaps the prettiest numbering system of all" and also discusses an abortive Soviet effort in the direction of ternary computing:

http://web.archive.org/web/20011205185830/http://americansci...

In an aside, the article hints that e-nary digits (base 2.718…) if somehow made practical/meaningful, might actually be better than ternary (or perhaps even optimal?).

So maybe this paper's observation that ~"1.58 bits" (ln2(3) binary-digits) is a sweet-spot could be further refined into some method for representing the state of a e-nary-modeled algorithm in ln2(e) binary-digits (~"1.44 bits") per underlying e-it.

(As it may be of renewed interest, I've also put this 2001 "American Scientist" base-3 intro as a new HN submission for discussion: https://news.ycombinator.com/item?id=39541756)

gojomo··on Apollo calls AI a 'bubble' worse than even the dotcom era
'Bubbles' and 'gold rushes' are different things.
gojomo··on Press Release: Future Software Should Be Memory Safe
Sheer cost of "hundreds of billions"?

Within another couple years, automated systems that are descendants of today's rapidly-advancing coding-assistant LLMs will be able to do amazing things.

For example, it's plausible they'd be able to to create proven-memory-safe rewrites of older software, with verified drop-in compatibility for all enumerable use-cases, practically for free.

Well, unless it's a parasitic government contractor doing the work.

gojomo··on Ending the Ivy League's Tax Dodge
Would be hilarious to see Harvard & MIT relocate (in part or in full) to a friendlier tax jurisdiction – New Hampshire? SBF's Bahamas? – to avoid or lessen this tax.
gojomo··on Signal v7.0.0 with phone number privacy
Signal or any surveillor surrounding their servers (with or without Signal's cooperation) almost certainly has enough timing/traffic-shape info to reconstruct who-to-who logs.

"Sealed sender" (and some of Signal's other tactics) just demonstrate: Signal's main & disclosed codepaths aren't stockpiling the canonical metadata via the same blatant & undenied mechanisms of other services. Sufficiently sophisticated outside attackers, or insider threats, can construct nearly-equivalent logs via other means. (And: Signal seems reluctant to make choices, like truly ditching phone numbers as account IDs, that could limit these 'shadow' leaks.)

gojomo··on Permutation City (1994)
If they ever want to shoot an 'Even Blacker Mirror' TV anthology, they should adapt Egan's short stories. 'Axiomatic', especially.

Two stories in that collection from before the modern social internet (1992), with non-internet somewhat fantastic premises, nonetheless often come to mind when observing modern online self-presentational & affiliational dynamics:

• 'The Hundred Light-Year Diary' - a method of receiving tiny (tweet-like!) messages from the future – eventually rationed out to all people! – examines questions of free-will & (self-)deception, at many levels

• 'Unstable Orbits in the Space of Lies' - how much of what you believe/aver is imposed by your neighborhood?

gojomo··on Over 2 percent of the US's electricity generation now goes to Bitcoin
Notably: the article text, and referenced source (US Energy Information Administration), contradict the headline's claim.

In the text, we see the "over 2%" (specifically "2.3%") number is based on a survey and the conjecture "if [those facilities were] running all-out".

But they don't run all-out!

The article itself mentions facilities that shut down when grids under peak load.

But also, if you click through to the linked US EIA study the article reports about – https://www.eia.gov/todayinenergy/detail.php?id=61364 – you get the following verbatim undistorted information:

"Our preliminary estimates suggest that annual electricity use from cryptocurrency mining probably represents from 0.6% to 2.3% of U.S. electricity consumption." – that is: not necessarily or truly "over 2%" as implied by ArsTechnica's deceptive headline & spin. Perhaps as low as 0.6%!

"In addition, cryptocurrency miners in areas with fluctuating power prices have reduced their electricity use in response to periods of high prices in wholesale power markets, given the sensitivity of their operational profitability to electricity prices."

"In practice, cryptocurrency mining facilities frequently run at less than their maximum designed capability."

Bitcoin mining is specifically drawn to just those times & places where electricity is especially cheap, and stops on a dime the moment rates surge over what's mining-profitable, due to other peaking demand or sagging supply. Thus the strained assumption of always-running-full that's required for the "over 2%" headline is false.

The EIA reports they "lack the data to provide a well-sourced estimate" of actual utilization, but at one point hand-waves a guess "80%". I suspect it's even lower, but even at EIA guess: 80% of 2.3% is less than, not "over", 2%.

Ultimately, though any negative externalities of Bitcoin's energy consumption can and should be addressed by adding our best estimates of the cost of such externalities into the electric rates paid by miners – & all other electricity consumers. That solves not just Bitcoin's externalities – but those of every other energy user.

Further, the way proof-of-work mining works, & difficulty adjusts, forces only those with the lowest-cost electricity to be profitable, in the long run. Difficulty rises to erase the profitability of everyone else.

And in any sane energy system, the lowest-cost electricity will tend to be from sagging demand at mismatched-generation places & times: excess wind/nuclear/hydro power at night, excess solar power at peak daylight. That is: electricity that isn't (& for the moment couldn't) be used for other things, and is even sometimes sold at negative prices when grid/generation stability requires someone to take the load.

In an ideal & rational generation grid, these regional/diurnal cycles could mean tons of electricity – far more than 2% of annual usage, by raw KWh, but not price paid – could be soaked up with countercyclical users like Bitcoin mining. And in so helping with utilities' fixed-costs & demand-smoothing, that consumption would primarily benefit other grid users' costs and reliability. It also provides de facto subsidies to cleaner-but-spikier generation (wind/solar) or cleaner-but-harder-to-adjust generation (nuclear/hydro) – relative to traditional fossil-fuels.

That is, no surprisingly high "X%" (X < 50) of some superficially-disfavored electricity use should necessarily cause concern – it depends entirely on exactly when/where/how that X% is taken. Some uses, & generation, are "essentially free (or better)" with regard to any limited resources or externalities of concern.

gojomo··on Ask HN: Can the black border atop the HN mourning header link to who died?
I understand tradeoffs & 'technical reasons'. But, a web framework that makes specifying tiny useful HTML edits noticeably harder in code than similar changes would be in plain-text strikes me as defective. That's especially true for a broadly-useful 'global attribute' – supported on all HTML elements – like TITLE.

Looking at the precedents in code near what you've linked, for how possibly-arbitrary STYLE and CLASS and COLSPAN and CELLPADDING and WIDTH attributes are set, it's hard to see why it shouldn't be as easy as something like:

  ; (tr (td bgcolor black vspace 5 title "RIP Name (YYYY-YYYY)"))
Of course, it might not yet be that easy.

But if that's hard to support, with a couple low-risk broadly-beneficial lines elsewhere, then rumors of Arc's concise expressive power seem exaggerated.

gojomo··on Expensive fridges are dying, leading to fraud claims
Are there any residential refrigerators that offer that?

Seems that'd mainly be desirable in uncomfortably warm climates. In cooler spots, a refrigerator's waste heat may be a welcome addition to indoor heating.

gojomo··on Ask HN: Can the black border atop the HN mourning header link to who died?
I can understand remaining minimal, and not shuttling lookie-loos to an active thread.

But remaining mysterious kinda works against the "memorial" aspect, and making the editorial decision "this passing is worth noting". It creates an "in-the-know"/"not-in-the-know" division that's somewhat against the ethos of being fair & open with those who don't yet know something.

A simple, "RIP, [name] [YYYY]-[YYYY]" would be a kindness to readers, and enhance the respect paid the deceased, while avoiding the thread clickthroughs you're concerned about.

And otherwise, we'll keep having these threads every time the honoree's identity is sufficiently obscure, and any associated stories/remembrances scroll off the front page.

gojomo··on Ask HN: Can the black border atop the HN mourning header link to who died?
That's currently false - no mention is on the 1st 2 pages. So another suggestion: keep the death news on the front page for as long as the black bar is up.

Otherwise the bar becomes a kind of unfriendly cryptic ingroup signal – "sheesh, don't you already know?".

gojomo··on The Curious Case of MD5
Nope. 2^64 is the number of tries needed for a 50% chance of a collision between two of the tries. You don't have to stay poorly read on this, and groundlessly allege wrongness:

https://auth0.com/blog/amp/birthday-attacks-collisions-and-p...

gojomo··on The Curious Case of MD5
Conventional analysis of non-contrived files would suggest only a 50% chance of an MD5 collision among 2^64 (18 quintillion) files.

So any SO account of "hash collisions in MD5 (128 bit) after hashing around 20k files" may be indicative of some other bug, misreporting, or having a set of files that includes pairs intentionally-contrived to include MD5 collisions. (That's now trivial as long as you're not targeting a specific MD5 value, just intending two files to match, and have some ranges in the files where you can stuff the right result-aligning binary patterns.)

gojomo··on The Curious Case of MD5
MD5 is already not "fine" or "safe, even" against malicious actors who might pre-prepare collisions, or pre-seed their documents with the special constructs that make MD5 manipulable to collision-attacks.

Even if your extrapolative method was sound, you've already got several factors wrong. The best SHA256 Bitcoin miners are today more than twice your estimate in hashrate, and on plain CPUs SHA256 is more like 4x slower than MD5. (Your smaller estimate of MD5's speed advantage is likely derived from benchmarks where there's special hardware support for SHA256, but not MD5, as common in modern processors.)

But it's also categorically wrong to think the CPU ratio is a good guide to how hardware optimizations would fare for MD5. The leading Bitcoin miners already use a (patented!) extra 'ASICBoost' optimization to eke out extra parallelized SHA256 tests, for that use-case, based on the internals of the algorithm. As a smaller, simpler algorithm – also with various extra weaknesses! – there's no telling how many times faster dedicated MD5 hardware, either for generically calculating hashes or with special adaptations for collision-search – might run, with similar at-the-gates, on-the-die cleverness.

Further, attacks only get better & theory breakthroughs continue. Since MD5 is already discredited amongst academics & serious-value-at-risk applications – and has been since 1994, when expert cryptographers began recommending against its use in new work – there's not much above-ground scholarly/commercial activity refining attacks further. The glory & gold has mostly moved elsewhere.

But taking solace in the illusory lack-of-attacks from that situation is foolhardy, as is pronouncing, without reasoning, that it's "probably decades or more" before second-preimage attacks are practical. Many thought that with regard to collision attacks versus SHA1 – but then the 1st collision arrived in 2017 & now they're cheap.

You can't linear-extrapolate the lifetime of a failed, long-disrecommended cryptographic hash that's already failed in numerous of its original design goals. Like a bridge built with faulty math or tainted steel, it might collapse tomorrow, or 20 years from now. Groups in secret may already have practical attacks – this sort of info has large private value! – waiting for the right time to exploit, or currently only exploiting in ways that don't reveal their private capability.

You are right that there's no present 'crisis'. But it could arrive tomorrow, causing a chaotic mad-dash to fix, putting all sorts of legal cases/convictions/judgements in doubt. Evidentiary systems should be providing robust authentication/provenance continuity across decades, as that's how long cases continue, or even centuries, for related historical/policy/law issues to play out.

Good engineers won't wait for a crisis to fix a longstanding fragility in socially-important systems, or deploy motivated-reasoning wishful-thinking napkin-estimates to rationalize indefinite inaction.

gojomo··on The Curious Case of MD5
Second preimage attacks aren't the only threat in a forensics environment.

Also, hand-wavy extrapolations from Bitcoin miners aren't a reliable estimate of how fast & energy-efficient dedicated MD5 hardware could become.

gojomo··on 1D Pac-Man
Up-down instead of right-left would still be a "single tunnel", "no off-shoots", "1D", "can't turn", etc.
gojomo··on 1D Pac-Man
Another thought: have the 1D RGB strip spin around its long-axis to enable rapid LED cycling to stroboscopically plot the domain in 2D, despite the action & rendering array being "only 1D".
gojomo··on TV Pixx – the benefits of screaming at your TV to win prizes
The `1D Pac Man' game reminded me of this curious chapter of videogames-on-TV history. Here's a better look at some of the alternate (bowling, baseball, space dogfight) variants that were offered: https://www.youtube.com/watch?v=UJN9eM84Rq8
gojomo··on 1D Pac-Man
Alternatively, the one dimension of movement could be up-down.
gojomo··on 1D Pac-Man
Your comment reminded me…

Long ago – early 80s – UHF & cable-syndicated TV station WPIX (NY) offered live broadcast call-in videogames controlled by the caller's voice yelling "pixx!". Here's a peek of some of the forms it'd take:

https://www.youtube.com/watch?v=UJN9eM84Rq8

"Behind the scenes, the voice activation was really operators in the control room pushing the buttons." :)

The simple control of this 1D PAKU-PAKU game (or others you're considering) could be wholly voice-operated. (A fun trigger word nodding to an old 'Friends' episode might be: "Pivot!")

With modern voice-recognition/voice-printing, it might even be possible to have a multiplayer game where even though N players are all yelling the same trigger-word, the recognition disambiguates the speaker to control only their sprite. Or, use a larger vocabulary of control-words for more actions. 'Nibbler' with yelled commands "Up!"/"Down!"/"Left!"/"Right!" might be diabolically frustrating.

gojomo··on Show HN: Vowl daily word guessing game
Was eager to try but found it buggy - or at least, confusingly arcane in how guesses are scored.

Today's 'transportation' puzzle had 4 consonants & 4 vowels. I got to a guess (nonsense string) whose feedback was 4 vowels correct, 4 consonants correct. (Not sharing here to avoid spoiling for others.)

After becoming frustrated that I couldn't devise a word using those letters, I entered them into online word-finders like those for Scrabble or the Internet Anagram Server – and they're not turning up any single 8-letter word/words (transportation related or not, compound or not) that might be the answer.

Maybe brands? No luck.

Show to wife, she makes suggestion that's similar to the 8 letters I've been trying, but differs in one vowel. Bingo.

So: it was telling me I had 4 vowels correct even when I was supplying a vowel NOT present in the answer.

(A rot13 of a string it erroneously told me had 4 vowels and 4 consonants correct: ZNEBHVYA)

gojomo··on If only someone told me this before my first startup
If shipping something requires a 'team' – coordination between specialists (or worse, multiple 'owners') – it'll be harder than when each cross-cutting feature can be done by a broadly-capable/empowered 'full stack' person.

(At least, that's how I interpret the OP, and don't understand your perception of any tension between 'full stackers best' & 'teams unproductive'.)

gojomo··on Show HN: Chrome Reaper
Brave's shields against ad & tracking inserts seem to help against many of the cycle-stealers that affect standard Chrome… except for the occasional times where they perhaps trigger cycle-consumptive bugs by breaking assumptions of remaining JS.

I'd really prefer a browser to give a good estimate of an individual tab's perceptible load/contribution-to-lag – via some blended measure of CPU/interrupts/wakes/events – and give me a ranked, stable view of offenders, with an easy kill/reload/'nice' control.

The Chrome et al 'Task Manager' has only scraps of what's needed - with its noisy, jumpy process-centric view that hides obscures the load of individual tabs, offers only process kills rather than per-tab management, and often foregrounds internal processes that the user can hardly do anything about (or map to the truly-responsible web content).

gojomo··on Word2Vec received 'strong reject' four times at ICLR2013
As I understand it, no one has come close to the claimed results in s3.1 ("Sentiment Analysis with the Stanford Sentiment Treebank Dataset") and people have come closer but still not matched those in s3.2 ("Beyond One Sentence: Sentiment Analysis with IMDB dataset").

A thread where Mikolov is trying to help other with his patch to `word2vec.c` demoing a tiny bit of 'Paragraph Vector' – but reaches the limits of what he understands Le to have done – is: https://groups.google.com/g/word2vec-toolkit/c/Q49FIrNOQRo/m...

My own frustration (& reco to avoid the thin stanfordSentimentTreebank/RottenTomatoes data/results) is mentioned at: https://groups.google.com/g/word2vec-toolkit/c/ubFrO0a9Pe8/m...

I'd say that "concatenating PV-DBOW and (plain averaging) PV-DM" never seems to offer much lift compared the favorable way it's described in the paper and other Le comments. And after spending a bunch of time implementing the "PV-DM with concatenation (rather than sum/average) of many word-vectors as the context", as I interpret the paper's description "To predict the 8-th word, we concatenate the paragraph vectors and 7 word vectors", I've only seen it massively increase model size & training time for very little advantage.

Reddit searches on related topics turn up other anecdotes/resentments; eg:

https://www.reddit.com/r/MachineLearning/comments/18jzxpf/co...

https://www.reddit.com/r/MachineLearning/comments/hkiyir/com...

At a certain level, with so much "gold in them thar hills" in followup work – from both academic & commercial perspectives, I don't blame Le for rushing forward to other related fertile ideas, & ignoring the requests-for-explanation.

But there's something sloppy or fishy (hiding secret tweaks?) in the originally-claimed PV results, which has wasted a lot of time among those trying to understand & reproduce.

gojomo··on Unbricking my MacBook took an email to Tim Cook
Seems another example of Apple pretending it can't (or never would) do something related to stolen equipment – that they can & actually do if you just reach the right person with the right request.

My related story of someone likely circumventing my activation lock, using either fake docs or a compromised Apple-authorized agent, a few days ago: https://news.ycombinator.com/item?id=38622248

gojomo··on Apple to halt Apple Watch Series 9 and Ultra 2 sales in the US this week
Expert patent language typically covers every foreseeable usable variant, and any subset of features likely to provide the core benefits, as 'claims'.
← PreviousPage 7 of 34Next →