Back for another burble with apologies for the Necro. Probably best to keep my rambling in one place.
The company, their CEO and new consultant, presented at NEMS25, Bitcoin Park, at the end of January this year. A copy of the audio is available here,
https://podcasts.apple.com/us/podcast/nems25-reality-or-science-fiction-a-future-of/id1646515985?i=1000699494440Jose Rios appears to spend most of his time reminiscing about his time with Intel. Francesco Gardin does not manage to give much additional insight as to where the company is at. He does however drop some possibly interesting information subject to interpretation which might be a bit tricky.
Initially he suggests that their firstl AI models were trained assuming full access to the block headers, those parts that are subject to modification.
The work had to be thrown away because that was/is not the way ASICs are implemented. Perhaps related he says that Bitmain ASIcs do not roll the full nonce range. He suggests that they miss 30% of them. There is information about that here,
https://www.reddit.com/r/Bitcoin/comments/adddja/the_weird_nonce_pattern/Elsewhere Skot, of Bitaxe fame, has suggested that the anomaly arises because with a fixed number of ASICs within a rig and division of the work, nonce, across those ASICs with a requirement to synchronize that work rounding errors mean that some is missed. I'm not sure that would end up with 30% being missed but as a design decision it seems to make sense.
I'm not certain about the history or details but it seems that as initially proposed a 32 bit nonce in conjunction with the initial difficulty meant that solutions below the difficulty could be found. Of course as the difficulty has increased over time that is no longer the case and something else had to be added to the roll. Again being a knuckle dragger and not knowing the full details this appears to have been incorporated in the 'coin base'.
Whilst I m suspicious about the 30% figure mentioned by Francesco Gardin that particular number causes my mind to wander off in another direction.
The company has claimed that their AI models have been trained using historic blockchain data. Depending on the epoch then that data is likely to have been poisoned by Bitmain rigs, the market leader, throwing away 30% of the nonces. AI is very good at spotting patterns and Francesco Gardin has often claimed that his AI has spotted patterns in SHA256.
In particular Method C is claimed to be 30% better by throwing away inputs it thinks will not achieve a winning hash. 30% is a bit glaring because if you train your AI on data missing 30% of the inputs because the Bitmain ASICs threw it away your AI might end up thinking it has found its pattern.
A pattern that does not exist in SHA256 but was a result of design choices. You are not identifying a pattern in SHA256 you have come up with a Bitmain ASIC detector.
It is also worthless because you are throwing away the numbers the ASIC would have thrown away anyway.
A similar argument might apply to the claims for Method B. In this case Method B rather than throwing away inputs that will not win it supposedly selects inputs more likely to win. The claim is that it is 160% better. With the caveat that my maths might be broken 1/(1 - 0.3) = 143% OK we are not hitting things exactly but I will call that near enough. Method B is also a Bitmain ASIC detector..
Regarding Method A then things are harder to pin down.
It is only 10% better. However words from the company via PRN suggest that the company were only able to properly develop the AI for its model when they gained proper access to a Bitmain ASIC by using ESPMiner and this gave them the capability to perform more precise timing on how fast a particular input would hash. The implication would be that it has not been trained on historic data but it has been trained against the ASIC.
Assuming the ASIC, as a result of design decisions, is still throwing away nonces will hash the nonces it is throwing away faster because it is not hashing them. That one might qualify as a tautology but I'll rescue 10% not being 30% by suggesting the pool of data is less extensive than the historical data used to train the other models. Hence the discrepancy.
Just to mention I made reference to PRNs from the company when discussing Model A. The information is there but I will be damned, for the moment, if I am going to trawl through the inconsistent and indecipherable dribble that the company manages to publish via supposedly regulatory channels. I'll leave that one until they get upset with me but you can immerse yourselves in it here.
https://quantumblockchaintechnologies.co.uk/investor-relations/regulatory-newsFranceso Gardin is at pains to be proud of the supposed fact that the three teams who developed these AI models did so independently. Now I have to trogg through more drivel...
https://www.youtube.com/channel/UCYNcH8Fv9wZ31ZOeQeAmhwQhttps://www.youtube.com/watch?v=5ehU3hDMFEMThis premise falls flat on its face when you realize that these three independent teams trained their models on the same data. Again that one is in the PRNs. It's just piss poor Science. Francesco will/has also claimed that he has 2 petabytes of self generated data that was also used for training. Again three teams independently training their models on the same data. As an aside ask yourselves how quickly the Bitcoin network would fill up 2 petabytes of storage with wrong answers?
My guess is about 20uS give or take a (couple of) order(s) of magnitude.
We also get this one,
https://www.youtube.com/@Proactive247/search?query=quantum%20blockchainhttps://www.youtube.com/watch?v=p4UdQh9JCRcMethod A is now only 3% better, Previously 10%, and Method B is now only north of 10%. Previously 160%. Supposedly because they made things more generic. I refer to these interviews as being Proactive Handjobs and they come with a warning from the producers that the company taking part might be talking their own book.
https://www.youtube.com/watch?v=uXoDWjrUSZAThe above is prior to the presentation at NEMS25. Francesco Gardin describes it as being,..
"Now, where do we go from here? It's very simple. I mean we will be in the United States next week, at a very selective invitation only meeting with the top experts of Bitcoin mining and we will give an official presentation on our method. And that would be the beginning of our commercial engagement with the obvious players that have developed ASICS and they can add our IP to their existing ASIC chip so that the chip can boost their performance."
No offence to the crew at Bitcoin Park but Mr Gardin is bigging you up to take money off of others with less knowledge of your space in or outside of a hot tub.
On that point and in parting on April 8th the company claimed that they were in talks about Method C with an ASIC manufacturer,
https://quantumblockchaintechnologies.co.uk/images/QBT_-_NDA_8_April_25_final.pdf"The Company will be giving an in-depth presentation of the Method C AI Oracle technology to the manufacturer during April, followed by a period of testing by the manufacturer, to confirm the AI Oracle’s performance on their ASIC chip’s architecture."
Four months later that one has been memory holed the substitute Squirrels being,
https://quantumblockchaintechnologies.co.uk/images/QBT_-_Business_Update_19_June_2025.pdfTwo third party control board manufacturers and two new ASIC chip manufacturers. Apparently the products are available for immediate release and sales and porting of Method A and B to these third party boards involve no technical barriers, it's taken a month so far.. As is the case with the announcement of 8th of April I have no doubt that NDA's are in place.
Excuse me whilst I go off the wall but if any company is involved with such a process and feels that QBT technology does not live up to its promises I would suggest that they ignore the NDAs in place and advise the wider community of their findings. You might feel that you can walk away having dodged the bullet but the company will continue to raise money on the AIM market from private investors who will continue to drink the Kool Aid and dream of being millionaires.
Also I need to get a life which might involve laying out a board for my direct off-line regenerative 3kW Class D Audio Amplifier. No you cannot put one in your car to listen to fart bass at ear destroying levels. I can also do snake oil but I have standards.
