HNHacker News
TopNewBestAskShowJobs

ckemere

1,432 karma · joined April 21, 2017

Professor of Electrical and Computer Engineering {Machine Learning, Neuroengineering Technology, Learning and Memory}
submissionscomments
ckemere··on Knowledge and memory
As a memory neuroscientist, I enjoyed the shoutout here to episodic memory. It strikes me, however, that a feature that I've noticed when observing "reasoning" models is that they may explicitly search for evidence for intermediate pieces of their responses. If we're following the "remember"/"know" distinction developed by Squire and others, perhaps the more apt analogy might be that a singly pass through an LLM is similar to a "I know this" result, prone to hallucination, conflation, etc., and the multipass reasoning model might be more akin to the "I remember this" result, where primary evidence serves as a substrate for the response?
ckemere··on Knowledge and memory
Addressing the second paragraph here - while conflation and reconsolidation are real phenomena, it's also quite clear that most humans form episodic memories. Some quite clearly have incredible abilities in this regard [2].

A distinction between semantic (facts/concepts) and episodic (specific experiences) declarative memories are fairly well established since at least the 1970s. That the latter is required to construct the former is also long posited, with reasonable evidence [1]. Similarly, there's a slightly more recent distinction between "recollecting" (i.e., similar to the author's "I can remember the event of learning this") and "knowing" (i.e., "I know this but don't remember why"), with differences in hypothesized recall mechanisms [3].

[1] https://www.science.org/doi/full/10.1126/science.277.5324.33... or many other reviews by Eichenbaum, Squire, Milner, etc

[2] https://youtu.be/hpTCZ-hO6iI?si=FeFv8MGmHTzkLd8p

[3] https://psycnet.apa.org/record/1995-42814-001

ckemere··on Neuralink 'Participant 1' says his life has changed
I think that the negativity here is unfortunate. The reality is that it’s very hard to see a normal VC level return on the $100M+ Elon and friends have invested here. And don’t let anyone fool you - this is the fundamental reason the BCI field has moved slowly.

If Neuralink proceeds to a scenario where quadriplegic patients can get reliable (ie lifelong) control of their computers for less than $100k that will be a huge win for them for a cost that no one else was willing to pay.

To be clear, at that order of magnitude they might make back their investment, but it won’t be 10x or 100x, and the potential healthy-brain-connected-to-the-AI play is much less rooted in reality than Teslas all becoming taxis.

Worst case scenario is that Elon loses interest and pulls the plug and Mr Arbaugh loses continued tech support a la a google product. I think that’s the one question I wish the author had asked…

ckemere··on The U.S. grid is so weak, the AI race may be over
Weren’t there Federal subsidies for wind? There certainly were/are philanthropic support (customers paying extra for “wind energy”). Unclear what role ERCOT played in these two factors. (Though in the current climate they might have tried to block them.)
ckemere··on New executive order puts all grants under political control
If there’s a specific researcher you want to fund, you can absolutely call them up. There will be paperwork around IP, independence, etc.

If you want them to do a specific experiment you’ll likely first need to convince them that it’s a good use of their time.

Harvard itself plays approximately zero role in the decision to do a study or not. (There are ethical oversight committees, etc.)

$1M total cost ($600-800k direct cost depending on terms) will buy you a postdoc’s time and effort for 5 years. Unfortunately most labs aren’t set up for a time/money tradeoff, so 5 postdocs for 1 year would be unlikely absent a really exciting project.

It definitely happens quite a bit!

ckemere··on ‘No Other Land’ consultant Awdah Hathaleen killed by Israeli settler
My point was that AFAIK, Palestinians would include Jews in the potential citizens of a Palestine. So that the conflict is not between two ethnonationalist groups.

There certainly is rhetoric around the ethnic origin of some Israeli citizens being Northern European rather than middle eastern, so perhaps the original claim has some validity.

ckemere··on ‘No Other Land’ consultant Awdah Hathaleen killed by Israeli settler
Many Palestinians identify as Orthodox Christian, including former PLO leaders. This is not solely a Muslim v Jew conflict AFAIK.

(Would love education to the contrary!!!)

ckemere··on ‘No Other Land’ consultant Awdah Hathaleen killed by Israeli settler
This is somewhat confusing. If there were no religious component, why not declare all Palestinians “Jewish” and resolve the issues immediately?

My understanding is that ethnically Jewish individuals (ie who can trace their matrilineal genealogy to synagogue records in accepted orgs) who have converted to other religions are not entitled to the right of return?

ckemere··on ‘No Other Land’ consultant Awdah Hathaleen killed by Israeli settler
It is unclear to me that Palestinians represent an “ethnonationalist” block. My understanding is that the definition of “Jewish” for the purposes of immigration to Israel is a somewhat precise combination of genealogical or religious criteria. But I don’t know of an equivalent definition for Palestinian. Can someone educate me on this? In the absence of such a definition, it seems that the primary conflict is whether Palestine should be specifically designated as a home for all “Jews” or merely as a home for the people that currently live there.

Again - I’d love further education about legal definitions.

ckemere··on IDF officers ordered to fire at unarmed crowds near Gaza food distribution sites
I upvoted this article because it reminds me that there are Israelis who are opposed to what is going on. And newspapers that take risk to report on it. And maybe even military investigators trying to stop it.

It’s no different with the American military.

ckemere··on U.S. bombs Iranian nuclear sites
I think the point about India/ Pakistan is a great counter. At what point would a nuclear Ukraine launch?
ckemere··on U.S. bombs Iranian nuclear sites
During the period when Israel was rapidly achieving air superiority I was reflecting on the hypothetical- “What if Iran already had a small number - 1 or 2 - poorly tested weapons. When would they decide to use them? Would the Ayatollah order them to destroy Tel Aviv in revenge for killing him?”

And now, assuming this sets their program back 10-20 years, what will the answer to that question be?

Has anyone seen any analysis on this? It feels like maybe the nuke might be less of a deterrent than threatening to acquire one because using it would put you in a MAD situation where your ballistic missle might malfunction???

ckemere··on Drones will realize the promise of suicide terrorism
Explosives have to be manufactured, and most require precursors that are hard to come by. More broadly, perhaps the historic argument that we should care about, eg Somalia, because failed states can enable chaos that crosses borders becomes more true???

(*edit - I mean by hosting factories for illicit materials!)

ckemere··on College English majors can't read
I guess my point is that maybe there is a conflation here of breadth of vocabulary and comprehension. Does my inability to understand a Drake dis track mean that I’m illiterate?
ckemere··on College English majors can't read
I can’t read Les Miserable in its original French (despite 5 years of high school French class), but I love it in translation. Does that make me functionally illiterate?
ckemere··on Trump administration halts Harvard's ability to enroll international students
I think in context “legacy” refers to the affirmative action boost given to children of (donating) alumni over better qualified unconnected peers.
ckemere··on Trump administration halts Harvard's ability to enroll international students
This is an explanation from my department chair which I've expanded. In the context of a university, there are four main power groups - the alumni, the faculty, the students, and the board of trustees. (Within each group of course are subfactions.) The actual power balance between these groups is never precisely certain (it's an unobservable "latent variable"). Whenever large events happen that involve the university, we get observations that allow us to estimate the latent variable better.

In the case of Harvard, I think the current observations are most consistent with the following: the Board of Trustees, faculty, and students have currently aligned in their goals - which we might summarize as (1) maintaining independence from the government and (2) the ability to hold/teach specific "controversial" viewpoints (benefits of diversity, anti-colonialism, potentially other "progressive" concepts). I suspect that within the factions the relative importance of these two goals is not balanced. The fact that the coalition has survived much longer than, e.g., Columbia, is somewhat surprising.

My suspicion is that the answer to your question is that the persistent "smacking around" is only in part due to the external factors other replies have mentioned. I think a major piece of the situation can be explained by a change in the power dynamic with the alumni. Under normal circumstances, the faculty presumably hope to maintain long lasting influence over their alumni, which the board of trustees leverage to bring in more money and influence to the university. The current situation suggests that the high-power/high-$$$ portion of the alumni who are in a position to leverage the public conversation about what's going on are not doing it. This implies that the strength of that edge of the power graph is much weaker than it was expected to be. I think it remains to be seen whether this is true. Further observations that would support that would be reduced donations, public complaints, etc. Conversely, increased fundraising and more public support would suggest the opposite.

The key point about the university power network is that USUALLY, the best situation is to avoid situations that actually reveal too much information. Everyone would prefer to believe they have more power than they do. Obviously the alumni are composed of factions, and presumably a large fraction of the potential participants are also members of other organizations with latent power networks and participating in this particular situation would involve expending capital in these other networks with potential reduction in power. Some alumni that have spoken up (i.e., Ackman) are clearly unaligned with the current coalition, and this MAY reflect the fact that the wealthy/powerful group of alumni that have sustained Harvard are really unhappy with the current stances of the university and would like it to shift (return?) to a different set of ideologies. But it's also possible that he represents a minority, and the rest are just nervous about getting involved.

My conclusion from this analysis is that things will persist as they have, with everyone who might be involved hoping that lawsuits will be successful in resolving the situation with the minimum of their involvement. If this approach is unsuccessful, I think we'll end up in a situation where we get a much better observation of the power balance between alumni, faculty, and board (I think the students rarely have as much power as they think they do!).

ckemere··on A Secret Trove of Rare Guitars Heads to the Met
The Met is a serious place. Apparently they developed a laser-based accelerated aging system so that they could measure (and subsequently monitor) how the finishes on these guitars change due to exposure. Also crazy environmental control in the context of millions of people visiting, with the goal of having the instruments preserved for posterity.

The curators are an unusual mix of artist, scientist and engineer. They all have PhDs, and apparently there’s a concept of tenure.

ckemere··on [dead]
I love(d) the Arcteryx Captive Polo shirt. Arcteryx manufactured them until 2022, producing two new colors each year in addition to a standard black. Despite the $80 price tag, I bought them steadily over the course of the 2000/2010's, until I amassed a collection of almost 20 in different colors. I still wear one almost every day. I realize that what I appreciated most about them was the way that the fabric was cut to be form fitting. By my count, they have 7 fabric panels, not including the collar. (Front, back, 2 arm pits, 2 shoulders, and the left chest.) Compare that to your typical polo shirt which usually just has two. I assume that's why Arcteryx discontinued them. (Insidiously, they brought back the name this past year for a standard 2-panel shirt.)

Can anyone recommend a service which can make shirts to order? Or alternatively, an AI shopping agent that is capable of recognizing the seam locations in shirts online so that I could find something similar from another vendor?

Thanks!

ckemere··on America's College Towns Go from Boom to Bust
Musing - Universities historically were easy to create in rural/underdeveloped areas with cheap land. (Particularly the ag/tech variant funded by the federal government.) A very small number aggregated enough people to generate sustained urbanization. The rest have always been tenuous in the sense that they exist because of a state-federal consensus that they are worth subsidizing. Without this, urban/suburban universities are more likely to sustain themselves.

Further thought - on the timescale of history, professors have always been (lower) middle class. Perhaps we should not be surprised to see a return to the mean…

ckemere··on Baby is healed with first personalized gene-editing treatment
The work outlined in the actual paper is quite remarkable. Within this 6 month period they generated transgenic mice with the precise mutations (paternal and maternal) this baby had in order to test treatment for functionality and toxicity as well as a doing a toxicity study in monkeys.

To those thinking about commercialization opportunities, these two steps seem the most labor intensive and time consuming, but also the most necessary in order to actually have confidence to inject completely customized gene editing therapy in a kid.

(Also worth highlighting for folks opposed to animal research.)

ckemere··on Neuralink Eyes Fundraise at $8.5B Valuation
(No access so haven't read yet.)

I think that the current systems being tested by Neuralink are (1) a connection to motor cortex allowing quadriplegic/locked-in to type/control a mouse; (2) a connection to visual cortex to provide vision for patients with bilateral complete vision loss.

(1) has been documented fairly well; provides some benefit (let's say 10x) for patients beyond eye trackers / sip-puff interfaces but at a cost that is easily 10000x greater. Unclear whether insurance would cover this. Even if they did, total patient population increase per year is probably ~1000, so asymptotic market is tiny.

(2) has not been documented. Historically, there are lot of challenges with cortical stimulation for vision, and the company's distaste for doing science / acknowledging scientific uncertainty makes me skeptical. Same issues as (1) apply if it does work.

I think that people who worry about black-mirror style issues have been overwhelmed by the hype machine. These things are infinitely worse than natural senses and motor control. They are using bluetooth connections and they have no R&D path to increasing bandwidth to neurons in a way that would grow exponentially.

What is possible in this space is something that has real but marginal benefit to a small number of patients. I had hoped that somehow Musk was motivated to solve the insurance and regulatory challenges towards deploying things to these patients, but I think unfortunately that the more their valuation rises, the greater the likelihood that they will close up shop sooner.

ckemere··on More Everything Forever
Given that no one has yet traveled to Mars, “faced a far more dangerous journey” seems a ridiculously hyperbolic statement. (Thinking even about the lost colony at Roanoke.)
ckemere··on Has anyone else found Google's AI overview to be oddly error prone?
Why offer a solution then? Seems fairly easy for google to avoid giving the final number?
ckemere··on Has anyone else found Google's AI overview to be oddly error prone?
Of course! I was originally imagining google would give me a website with an embedded calculator. I was most surprised by how everything was beautifully accurate up until the end when the number felt suspicious. (Much less suspicious than the examples I posted, actually.)
ckemere··on NSF halts grant awards while staff do second review
Hi Dang!

Thanks for the note. I definitely understand your reasoning. I will say that to me (a) finding out that NSF has blocked funding is a new and different point than that DOE has blocked overhead (or that NIH review panels were on hold). And (b) this news would not have reached me as soon if I hadn’t noticed it in the /active front page. And (c) that I come to the discussions partially in hope that I’ll learn more “secret” details about what is happening. (d) I recognize that much of the discussion is repetitive, but mixed in there is useful analysis and even more actual news reporting. (e) I understand that people get overwhelmed by the continued raging and conflate new news with seeming opinion. (f) I recognize that your job is hard but know that I really appreciate HN!

ckemere··on Dept of Energy cuts indirect rates to universities
While they claim to be seeking “efficiency” the truth becomes clear if you read carefully. They have restricted the class of grants to which this rule applies to “institutes of higher education”. Why not apply it to for profit companies also??
ckemere··on US Administration announces 34% tariffs on China, 20% on EU
OR use borrowing (e.g. current account deficit). If government spending drives productivity growth then it’s a net positive?

OR tax wealth. If most return on international capital investment is being stored in the US, taxing this effectively taxes profits on international sales???

ckemere··on US Administration announces 34% tariffs on China, 20% on EU
I read this and I was quite disappointed that he didn’t talk about labor costs and other comparative advantages for labor-heavy manufacturing to be in a different situation trade market-wise than it was in the 1930s.
ckemere··on US Administration announces 34% tariffs on China, 20% on EU
I’ve reflected that resource extraction jobs often end up in the high wage / low educational investment category, and I’ve wondered if that motivates the whole “annex Greenland” bit as much as “securing critical resources” does…
← PreviousPage 3 of 11Next →