dekhn · HackerLangs

l’altro ieridiscuss

Good point. I've ended interviews (as a candidate) when I believed the interviewer was being lazy (for example, administering a leetcode question and not even changing any of the details like input or output data). I ended up writing my own questions that aren't in leetcode because I interview candidates now. And I give 100% attention to the candidate.

dekhn

l’altro ieridiscuss

Literally the second I read "it's a data stream" I knew the answer was going to be reservoir sampling.

RS is really interesting to me. many people you talk to can realize you can compute the mean of a data stream (https://www.geeksforgeeks.org/web-tech/expression-for-mean-a...) without knowing the exact formulation. And it's not far from that to think of a sampling strategy to decide if a new sample should go into a fixed-size reservoir. (for all of these, I know specific hints that will usually help people get to the next step).

The only reason I know RS is because it was in the google3 monorepo and I was looking for interesting codes to use and found it. There was an associated Sharding class, LexicographicRangeSharding (https://www.mongodb.com/docs/manual/core/ranged-sharding/) which you could use to find near-optimal split points in sorted string tables so your mappers didn't end up with hotspots. If you had shown me Algorithm R in a stats class, I don't think I would have appreciated it at all, but seeing the code implementation and a useful example made it click.

dekhn

3 giorni fadiscuss

Did you actually find any actionable results? When I had my genome sequenced professionally they told me I literally had no SNPs in my genome that were associated with deleterious mutations. And even if you do, there are only some treatments- and those treatments usually have clinically validated tests that are accurate.

dekhn

3 giorni fadiscuss

.... which does not need or really want home sequencing. That's clinical-grade with QC and high quality instruments, along with professional genetic analysis.

dekhn

4 giorni fadiscuss

Nothing about this is the future. Sequencing at home will not solve any major problems. It's mainly a fun exercise to demonstrate that sequencing has been commodified.

dekhn

4 giorni fadiscuss

I'm of the general belief that something can be unaware and not have any understanding (in the subjective experience of consciousness), while also appearing to do so, and be useful.

That's what Bender doesn't get. Some folks tried techniques different from her favorite techniques and made unbelievably fast progress across a wide range of previously insoluble problems (regardless of whether they satisfy properties Bender believes are required for intelligent systems).

I think it's safe to say that none of the main LLMs have some sort of self-awareness as we think of it in humans, but I also expect that more sophisticated systems in the future could. If I had to guess, they would have significantly more activity going on in the network- not just individual end-to-end top-down forward graph, along with cycles instead of trees, and the neurons themselves would be sigifnificantly more capable (effectively little state machines that run functions on input that passes through). I guess also you'd want to have some sort of rules-based (but statistically trained) execution component managing everything.

dekhn

4 giorni fadiscuss

I mean, one of the most useful metal fabrication techniques is already folding thin sheets of metal; I learned it in shop class decades ago and it's still a very relevant skill. Some fun stuff here: https://www.youtube.com/watch?v=dS5kwdaNhZo

dekhn

4 giorni fadiscuss

Yes, they are. Likely due to a deep relationship between math and physics, statistical modelling of complex natural phenomena has repeatedly been shown to be the most effective approach. This is true of LLMs, but also of many stochastic (and other) systems.

dekhn

4 giorni fadiscuss

There's plenty of beautiful math there, but the relationship to what our neocortex does is pretty distant. Individual biological neurons can do fairly complicated things, including compute 10-bit parity functions (you would normally need a 3-layer MLP with a bunch of digital neurons to do this). And they don't seem to use backpropagation for learning.

dekhn

5 giorni fadiscuss

Statistical models have repeatedly shown themselves to be the most productive research method for working with complex human-based systems (and in the larger study of natural phenomena). It remains unclear whether there is any short term path for symbolic methods to catch up and exceed the capabilities of current/near-future statistical systems.

To me the real question begins only once we have a clear example of a non-trivial scientific discovery that is implicit (IE, not an obvious outcome of reading the literature and talking to the experts) and experimentally verifiable. Once that happens- especially if it is a reproducible process (IE, more discoveries) and it's significant (IE, impacts human life and mind in some profound way)- then the onus very much lies on Bender and her coauthors to explain whether we need more than a sufficiently advanced stochastic parrot.

dekhn

5 giorni fadiscuss

Personally, I've always read that paper as a political criticism of industry and industrialized research and capitalism. After decades in academic (and industrialized research) I've learned that smart people can write convincing takedowns of things they hate- and those takedowns, due to being well written, often punch above their weight in terms of impact on the community.

I think this paper would have been best split off from the conjoined criticism of environmental effects (which could have been its own paper, but not one published by Google, since their leadership's fundamental beliefs disagree with the paper's environmental impact premise. And the remaining part on text models could have been a bit more focused on the technical issues associated with statistical text processing and meaning, rather than criticism of the power structure that is loosely associated with the current AI push.

dekhn

15 giorni fadiscuss

Most of their fabs were divested to GlobalFoundries, but they still have pretty significant fab capability and capacity- I suspect at least partly to have a us-based chip-making for military ("Trusted Foundry").

dekhn

15 giorni fadiscuss

The author list will show "M.D." instead of (or in addition to) "Ph.D.". The paper terminology will be skewed towards medical terms.

dekhn

15 giorni fadiscuss

I've seen this a few times- for example, my manager at Berkeley Lab was a PI but only had an undergrad degree. This is very rare, however.

dekhn

16 giorni fadiscuss

If I type your first query into Gemini, it immediately spits out a long and plausible answer.

What exactly are you saying it's refusing? Can you give a screenshot or example?

dekhn

16 giorni fadiscuss

then there is some detail missing from your report- for example, if leadership asked you to take down the repo (perhaps because another team was building/releasing/launching their own) and you refused, then you weren't fired for creating the Google workspace CLI, but for something else.

In short, the way the post is framed, and the reality behind it, don't seem to match up, and people with experience are asking for you to clarify. If you don't want to go into more detail, that's fine, but... many people (like me) read what you wrote and thought "there must be more detail than this, because it would be silly for somebody to get fired for doing their job in good faith, and following the rules"

dekhn

16 giorni fadiscuss

I'm not 100% sure what your point is here, but the point I was making is that even in theory if you have a "firewall", your contract with Google (that you signed and agreed to) says that Google owns your code output even if it's done at home on your own time. It really comes down to (at least for Googlers in California) whether the work is related to Google's business (which often covers things like games, even if Google is not a game developer), and even then, you need to submit the work to the Googler lawyers and have them make the determination, before you get permission to publish the code on your own.

I saw this play out repeatedly with OSPO and saw many people who believed they had set up firewalls to allow this, and I understood Google's position and appreciated that they made it fairly easy for most people to open source their code.

dekhn

16 giorni fadiscuss

The Register is a tech paper that is modelled on various British tabloids (daily mail, the sun). Sometimes it's humor, sometimes it's real news and occasionally they even break a new story.

dekhn

16 giorni fadiscuss

how did you get from "raising your voice in a meeting" to "toxic assholes" and "racist slurs". Those are very different.

dekhn

17 giorni fadiscuss

Google has multiple orgs on github: google, google-cloud-platform, chromium, android, flutter, angular, tensorflow all have their own top-level orgs because google ships its org chart (https://en.wikipedia.org/wiki/Conway%27s_law). Some orgs have been created by google and then released to the wild (kubernetes).

I think but I'm not sure that this is a "semi-official" org run by Google DevRel. Perhaps it has looser rules and ownership than the more official orgs? If I'm using the Wayback Machine properly, https://web.archive.org/web/20201130062102/https://github.co... shows that the site already used the Google logo way back in 2020 (earliest snapshot).

comments