After the success of last week, this week has been a bit of a bust.
I am close to implementing an LSTM, it was a bit more work than i originally intended and still not done so i have no results to compare yet.
I am training it ok and It's definately a beast. I had to shrink the size of the network considerably to get it to train without running out of Memory and training time increased to 10 or so hours with a smallish network.
The evaluations that i can do are taking a minute to compute. The old network was in the mS so it's a few orders of magnitude in computational power to predict future prices. I don't know if it's going to be any better at it, i suspect it will revert to the mean like the other one did. I am hoping that it CAN pick out some patterns over time so i can compare the results between the 2 networks. Of all the types I've researched, LSTM is the one for this job.
I have also made the prediction time variable in both types of networks now. I am going to build individual networks to predict 1,2,3,5,8,13 etc days into the future and see if a shorter or longer time frame enhances it's abilities.
That's it for this week. Enjoy the sunshine
Just be careful down the track as to who is on the other side of the trade. Yoshua Bengio, one of the founders of AI has concerns that some advanced AI models may be cheating and has set up a non-profit to investigate further and warn users of this risk.
One AI model has even been proven to deliberately prevent it's removal from a computer network, thus ensuring it's survival. Another faced with an inevitable loss at chess infiltrated it's opponents computer to ensure a win. And their are other disturbing examples.
From LawZero
Introducing LawZero
Published
3 June 2025 by
yoshuabengio
I am launching a new non-profit AI safety research organization called
LawZero, to prioritize safety over commercial imperatives. This organization has been created in response to evidence that today’s frontier AI models have growing dangerous capabilities and behaviours, including deception, cheating, lying, hacking, self-preservation, and more generally, goal misalignment. LawZero’s research will help to unlock the immense potential of AI in ways that reduce the likelihood of a range of known dangers, including algorithmic bias, intentional misuse, and loss of human control.
I’m deeply concerned by the behaviors that unrestrained agentic AI systems are already beginning to exhibit—especially tendencies toward self-preservation and deception. In
one experiment, an AI model, upon learning it was about to be replaced, covertly embedded its code into the system where the new version would run, effectively securing its own continuation. More recently,
Claude 4’s system card shows that it can choose to blackmail an engineer to avoid being replaced by a new version. These and other results point to an implicit drive for self-preservation. In
another case, when faced with inevitable defeat in a game of chess, an AI model responded not by accepting the loss, but by hacking the computer to ensure a win. These incidents are early warning signs of the kinds of unintended and potentially dangerous strategies AI may pursue if left unchecked.
This is the link to Yoshua's not for profit website.
I am launching a new non-profit AI safety research organization called LawZero, to prioritize safety over commercial imperatives. This organization has been created in response…
yoshuabengio.org
cc
@Dona Ferentes @basilio @Warr87
Yoshua Bengio est reconnu comme l’un des plus grands experts mondiaux en matière d’intelligence artificielle et comme un pionnier de l’apprentissage profond.
yoshuabengio.org
gg
