Great article as an outsider just to understand a bit what the open questions are in studies of the brain and how these open questions get discussed by researchers. Very easy to read and fascinating.
626 karma · joined April 8, 2022
I still hold that there is good reason to be concerned with researchers claiming to objectively evaluate a silicon based sentience and jumping the gun a bit (imo) by writing articles about LLMs which "dream" and "hallucinate" when the behaviour is a consequence of essentially a non-linear regression model that does not yet show clear signs of the high order emotion/reasoning/awareness/intention
Many who research LLMs avoid these terms in their publications for good reason (again imo)
''' “I could say — these models hallucinated; or, to be more precise, I could say, well, the model made an error, and we understand that these models make errors,” Fayyad says. '''
For years we called the behaviour that researchers now call "hallucination" test error or misprediction or a failure to forecast correct outputs based on "out-of-sample" inputs and those terms were accurate and would still be accurate to use on today's LLM models.
Metaphors are great but can be misleading, people don't often make an argument that a computer mouse can feel pain.
[1] https://news.northeastern.edu/2023/11/10/ai-chatbot-hallucin...
I am not an AI researcher, my field these days is much smaller statistical models, but reading some of these papers and then reading also the discourse a lot of AI researchers engage in surrounding AGI, sentience, pain, etc... I for one have a hard time taking some of it seriously. I just don't buy some of it but of course I'm less than fully informed.
My only critique is that you should not start by naming every algorithm or operation you will perform on a model after a thing that a person with a brain does and then claim to evaluate a concept such as sentience or distress from the position of an unbiased objective observer. This naming convention sows doubt among those of us who are not a part of that world.
> Feature visualization, also known as "dreaming", offers insights into vision models by optimizing the inputs to maximize a neuron's activation or other internal component. However, dreaming has not been successfully applied to language models because the input space is discrete.
Will you share the github source?
https://www.bostonglobe.com/2025/02/27/nation/rising-border-...
line from the archived website:
"The Haskells wanted Canadians and Americans to have equal access to the Library and Opera House and so they chose to build on the border. Construction began in 1901; the Opera House opened in 1904 and the Library in 1905."
But to be fair busses and trains and other public transit that does fixed routes every day is the perfect bench for any clean energy drive train that needs real world testing.
Also public money should never be overly shy of coordinating with companies who are trying to in good faith to solve our climate crisis.
Some discussion on this thread about hydrogen being a poison pill promoted by oil interests which I don’t really know about.
Electric bus programs have been broadly successful and no doubt have contributed in a real way to our understanding of that technology.
“ About a third of Tesla’s $35 billion in profits since 2014 has come from selling federal and state regulatory credits to other automakers. The credits are given to automakers that meet certain standards, including selling a certain percentage of zero-emission vehicles. Tesla is the largest seller of these credits to automakers that don’t meet the standards and want to avoid paying a fine.”
[1] https://www.thestar.com/business/tesla-sales-are-down-but-it... [2] https://www.iea.org/reports/global-ev-outlook-2024/trends-in...
Not to beat a dead horse but looking over what I believe is the patent [1] for the oxygen delivery device in question this seems to be another case of a standard industrial control algorithm being dressed up as "AI"
Ethics of this aside the above claim must be dubious I would think the majority of manufacturing inefficiencies are due to down time as a result of raw material shipping delays or machine break down… of course I’m in no position to offer an informed opinion but just based on the product website I have a hard time taking this stuff seriously.
Monitoring of factory workers isn’t hard to do with current surveillance and 1 or 2 humans in the loop
It makes learning new software much slower and basically requires going to the docs or a video tutorial if the ui design isn’t exactly spot-on and perfectly intuitive which it almost never will be
Is this what Hofstadter means by a strange-loop?
As in: "The hacker gained access to" "The hacker took charge of" "The hacker assumed authority over"
I have yet to see a thorough explanation of what specifically was hacked here anyhow
When you decentralize finance like this what becomes okay to do according to system rules is exactly what is possible to do according to system rules. We don't have humans in that loop anymore to enforce moral judgments about what constitutes unlawful theft (except for 1 or 2 rare "hard-forks" of various blockchains to reverse devastating transactions).
I feel bad for people who lose large volumes of cryptocurrency to malicious actors in the same way I feel bad for people who lose large volumes of real money to a casino.
It is 2025 now and we all know that anyone who can somehow get your private-key to whatever blockchain backed assets you have "owns" those assets just as much as you do and they are permitted to take them under the rules of the system so whatever you do do not lose that key.
There is no higher arbiter of justice in this space so use it at your own risk.
That’s a neat proof of concept.
https://arxiv.org/pdf/2501.06623
And I'll include this colourful quote:
"This is orders of magnitude larger than the largest nuclear explosion ever detonated, so this is not to be taken lightly."
I quickly read through and may have missed it but I do not see any mention of the timescale over which this would work just that it could sequester ~30 years worth of CO2 output.
It's a new level of public-but-divided space when we all start running client side software to reinforce our world view. I'm certainly not criticizing that chrome extension I think that is funny, but it's also something to think on.
However all “good” ADs are less effective than they could be when the population experiences a nearly unavoidable torrent of “not good” ADs.
People are savvy and broadly know the difference but on a subconscious level likely begin to perceive PSAs with more skepticism than is warranted when so much advertising masks itself as helpful advice when in reality it is selling a service based on nonsense statistics or bad faith argument.
I think some kind of pushback from society against the worst forms of advertising is needed.