Lightnews — Scholar-powered news

Rachel Flood Heaton

@rachelfloodheaton.bsky.social

1.3K followers 2.4K following 630 posts

Cognitive and perceptual psychologist, industrial designer, & electrical engineer. Assistant Professor of Industrial Design at University of Illinois Urbana-Champaign. I make neurally plausible bio-inspired computational process models of visual cognition.

Posts Media Videos Starter Packs

Pinned

Rachel Flood Heaton @rachelfloodheaton.bsky.social · Aug 22

In our forthcoming paper, John Hummel and I ask what it would mean for a neural computing architecture such as a brain to implement a symbol system, and the related question of what makes it difficult for them to do so, with an eye toward the differences between humans, animals, and ANNs.

From Basic Affordances to Symbolic Thought: A Computational Phylogenesis of Biological Intelligence

What is it about human brains that allows us to reason symbolically whereas most other animals cannot? There is evidence that dynamic binding, the ability to combine neurons into groups on the fly, is...

arxiv.org

1 13 34

Rachel Flood Heaton @rachelfloodheaton.bsky.social · 3d

Scenes from Chicago today

Rachel Flood Heaton @rachelfloodheaton.bsky.social · 4d

Some verifications are more independent and reliable than others. For example, I don't test my own code or review my own papers. I also don't test my own code or review my own papers and then just have someone visually check what I did. A different mind doing the stress testing is crucial.

Rachel Flood Heaton @rachelfloodheaton.bsky.social · 4d

Well, it depends upon the likelihood of an error in an activity as to whether the method is 'sound'. If one asks an unreliable system to generate the verification tool, and then visually checks the verification tool, one might be more likely to miss an error than if one wrote that tool oneself.

Rachel Flood Heaton @rachelfloodheaton.bsky.social · 4d

Although I submit that even a Fields medalist can make errors, so I wouldn't actually take the result as rock solid without some other sort of independent verification.

1 2

Rachel Flood Heaton @rachelfloodheaton.bsky.social · 4d

I am not a fan of current AI but even I acknowledge that it gives the right answers sometimes. The issue is that you have to already be an expert to know whether it was the right answer, and most users are not. So sometimes it causes legit tragedies by pushing naive people towards bad conclusions.

1 10

Rachel Flood Heaton @rachelfloodheaton.bsky.social · 6d

What replaced the program was subject specific tracking that can change dynamically and is based on long term classroom performance. I think it serves the same purpose with respect to pacing but does it more effectively.

Rachel Flood Heaton @rachelfloodheaton.bsky.social · 6d

It is amazing watching one set of students be treated like special geniuses and the other be treated like actual criminals when the difference might literally have been one point on a test that only some parents knew how to prepare their children for.

Rachel Flood Heaton @rachelfloodheaton.bsky.social · 6d

In my area, gifted placement was best predicted by having parental resources to train for the gifted testing. It perpetuated inequality, especially because there was no other onramp into the program past that initial testing. Kids who should have been considered 'gifted' were excluded too early.

Rachel Flood Heaton @rachelfloodheaton.bsky.social · 7d

I would be unhappy to be forced back to an internal combustion engine car. Riding in a car literally rattling from little explosions going off inside of it seems absurd now, like other ingenious but excessively complicated Victorian era contraptions. EVs are fast, clean, quiet, and easy to maintain.

Reposted by Rachel Flood Heaton

Kevin M. Kruse @kevinmkruse.bsky.social · 9d

Looks like Barney has finally had enough of Florida's bullshit.

88 370 2.8K

Rachel Flood Heaton @rachelfloodheaton.bsky.social · 8d

Having 'scratch' space to process pieces of information is useful across most contexts and types of computations. Even in humans this is true, like moving things in and out of working memory. However because this is so generally true, any true relationship to human-AI alignment is likely illusory.

Rachel Flood Heaton @rachelfloodheaton.bsky.social · 9d

The only thing I learned about antennas in my EE program is that antennas are weird and hard to design.

Rachel Flood Heaton @rachelfloodheaton.bsky.social · 12d

If people get this cancer I hope they are 74 years old.

Rachel Flood Heaton @rachelfloodheaton.bsky.social · 12d

This is starting to look more and more like some sort of fraud or market manipulation

Ed Zitron @edzitron.com · 12d

Premium newsletter: Based on my estimates and analysis, OpenAI needs one trillion dollars in the next four years to build 17GW of data centers and other commitments, with at least $500 billion needed for company operations. There is not enough capital to do this.

www.wheresyoured.at/openai-onetr...

OpenAI Needs A Trillion Dollars In The Next Four Years

Shortly before publishing this newsletter, I spoke with analyst Gil Luria, Managing Director and Analyst at D.A. Davidson, and asked him whether the capital was there to build the 17 Gigawatts of capa...

www.wheresyoured.at

1 5

Reposted by Rachel Flood Heaton

Ed Zitron @edzitron.com · 12d

OpenAI needs at least $500 billion in cash to fund its operations in the next 4 years, and $432.5 billion *on the low end* to meet their other obligations - more than the combined available capital of the top 10 PE firms ($477bn) and US venture capital ($164bn).
www.wheresyoured.at/openai-onetr...

OpenAI Will Need To Raise At Least $500 Billion Just To Fund Its Operations, And Another $432.5 Billion Or More To Fulfill Its Obligations - Which Is More Than The Available Capital Of The Top 10 Private Equity Firms and US Venture Capital Combined
So, I am repeating myself a little, but I really, really do not trust OpenAI’s revenue projections, and think they are at best unrealistic or, at worst, a deliberate attempt to mislead investors using the media.

Nevertheless, even if OpenAI is in a position where it is making $200 billion by the year 2030, it will need to have raised at least $500 Billion to get there, and when you add its obligations — including those given to NVIDIA — it will have to raise (or have partners raise) another $400 or so billion.

This is more than the projected available supply of US venture capital at the end of 2025 ($164 billion), and the combined available capital of the top 10 private equity funds ($477 billion), as well as the $50 billion of available direct lending capital from JPMorgan and Goldman Sachs (about $35 billion across two closes).

2 42 220

Reposted by Rachel Flood Heaton

Ed Zitron @edzitron.com · 12d

I cannot express how wild this quote is. This is a top tech analyst at a well-regarded firm saying that there is not enough capital to actually pay for all the stuff that NVIDIA and OpenAI has promised. It is so important that everybody realizes that *there is not enough money to do this*

Ed Zitron @edzitron.com · 12d

I spoke with analyst Gil Luria at D.A. Davidson, asking if the capital existed to build OpenAI's promised 17GW of data centers.

He said "of course there isn't enough capital for all of this," but "enough capital to do this for a at least a little while longer."

www.wheresyoured.at/openai-onetr...

70 1.3K 3.4K

Rachel Flood Heaton @rachelfloodheaton.bsky.social · 15d

It's an empirical question but it's possible that for obscure enough information it won't do a good job with that.

Reposted by Rachel Flood Heaton

Julia M. Rohrer @dingdingpeng.the100.ci · 16d

Currently attending a conference & our field is quite a bit into fancy modeling, so it’s time to repost this blog post.

Don’t try to squeeze your research question into whatever model is fashionable right now; try to build the right model for your research question.

www.the100.ci/2024/08/27/l...

Let’s do statistics the other way around

Summer in Berlin – the perfect time and place to explore the city, take a walk in the Görli, go skinny dipping in the Spree, attend an overcrowded, overheated conference symposium on cross-lagged pane...

www.the100.ci

8 19 89

Rachel Flood Heaton @rachelfloodheaton.bsky.social · 15d

There are a couple of issues with this approach. The first is actually preserving content accuracy, which is significantly less likely in more obscure subjects. The second issue is what happens when the material/training needs to be updated and no humans are bothering to write new textbooks.

1 1

Reposted by Rachel Flood Heaton

Zoe Drayson @zoedrayson.bsky.social · 16d

When a typo in your Google search leads to new and exciting philosophical positions

AI Overview: "Dental realism"refers to the philosophical debate about whether teeth exist independently of human perception, a concept rooted in metaphysical realism. While realism asserts an objective reality for teeth, idealist philosophies contend that teeth's existence is tied to our consciousness and perception. Philosophers use this idea, for example, to discuss the ethics of tooth extraction, examining whether a procedure impacts a real entity or merely a subjective experience.

56 400 1.4K

Rachel Flood Heaton @rachelfloodheaton.bsky.social · 18d

They're blending it up and disguising it as other foods like ranch and ice cream.

1 1

Rachel Flood Heaton @rachelfloodheaton.bsky.social · 18d

I have seen this trend

Reposted by Rachel Flood Heaton

Terry McGlynn @hormiga.bsky.social · 20d

I strolled in and it was a bit toasty. So I turned the thermostat down, and the room cooled off. Oh. My. God.

After living through 20 years of placebo thermostats in offices, classrooms, and labs, I'm flabbergasted.

3 1 21

Rachel Flood Heaton @rachelfloodheaton.bsky.social · 22d

We did have some requests from grad students to defer.

Rachel Flood Heaton @rachelfloodheaton.bsky.social · 22d

We don't agree that there is any real match whatsoever between LLMs and brains, especially not human ones.

Rachel Flood Heaton @rachelfloodheaton.bsky.social · Aug 22

From Basic Affordances to Symbolic Thought: A Computational Phylogenesis of Biological Intelligence

arxiv.org

1 2