This BLOG is for LLNL present and past employees, friends of LLNL and anyone impacted by the privatization of the Lab to express their opinions and expose the waste, wrongdoing and any kind of injustice against employees and taxpayers by LLNS/DOE/NNSA.
The opinions stated are personal opinions. Therefore,
The BLOG author may or may not agree with them before making the decision to post them.
THIS BLOG WILL NOT POST ANY MAGA PROPAGANDA OR ANY MISINFORMATION REGARDLESS OF SOURCE.
Comments not conforming to BLOG rules are deleted.
Blog author serves as a moderator.
For new topics or suggestions, email jlscoob5@gmail.com
Blog rules
Stay on topic.
No profanity, threatening language, pornography.
NO NAME CALLING.
No political debate.
Posts and comments are posted several times a day.
The first AI I was able to check by chance came mainly from an article I had seen the day before. There were three numbers in the paragraph, say a, b, c. b was four times a, c was four, and it was said that b=c×a with c=4. But the AI got it all wrong. It changed b so it was no longer 4a, kept a unchanged, but still said that b was 4 times a. Stupid. Recently I searched for an update on the Pike County, Ohio, murder cases. The AI search said the relevant county was in Missouri, said the trials had concluded when in reality one has not started, got two of the involved correct but did not mention the other two people including the pending trial, and one or two other mistakes, all in one not at all long paragraph. I have seen recently that for AI generated paragraphs there is often a disclaimer that there could be mistakes. DAH.
For several years I have used Google Scholar to find citations of papers. Recently the search was switched to AI. One of my papers was listed by AI with over 800 more hits than before. I would take this but it is incorrect and I have checked. I have recently read that AI for Google is very much slower than the previous methods. If you are using AI, check.
LANL sponsored a AI Jam session at a casino and they got close to a thousand people to ask Open AI questions and help it learn. The general sense was total disappoint on the part of the participant but Open AI got some good information for free. Many people said you simply cannot trust the AI for anything remotely complex. It also can screw up a lot of even even simple things when you had any kind of slight complexity to it. If you know the stuff you can find the issues and fix them. The real danger comes from people who do not know the stuff that well and will simply trust whatever the AI says as they have no way to check.
If just recognizing that three numbers should be related is considered as complex, then AI is pretty stupid and is useless except for trivia issues. Not at all surprised. Hype is ruling our world and bringing us down, not only in AI. When I retired, people asked me what I was going to do. I told them I was going to watch the decline of science at the Labs and the decline of the Western societies. The decline started somewhere around the Vietnam War era. It is happening, note drugs in our societies. And do not place the blame on the drug cartels. The drug situation is a classic case of suppy-demand. If there is demand, there will arise a supply. I did not know, and still do not, know the details of how things would play out. That is why I am watching. All is on track.
An example of hype that can be argued to mark the start of the decline of the US. The US government hyped that North Vietnam attached US ships in the Gulf of Tonkin. Congress used this hype to pass the Gulf of Tonkin Resolution, a thinly disguised declaration of war. Years later McNamara admitted in his memoir that the attach never happened.
A lot of this doomsday talk about LLMs being more dangerous then nukes or taking over the world is self serving. The companies are trying to make headlines and say, "Look at us, it's so powerful that it's dangerous!". Well, guess what, we could also hook a machine gun up to a computer and program it to shoot people. But we don't generally do that because it's known to be dangerous. Similarly, don't hook up an LLM to the Internet and let it have free reign. It's a machine, if it starts to malfunction, turn it off. Better yet, keep it in a sandbox. The ability to do stupid stuff with computer networks isn't exactly new.
6:19 -- I think it is good to understand history. But there were other wars started on false or inflated pretexts, for example the sinking of the Maine in Havana is now thought to be accidental, it was hyped to create the Spanish-American war.
It is interesting of course, that the Vietnam peace deal was preceded by the large Linebacker and Linebacker II attacks on Hanoi, to coerce the Vietnamese to the bargaining table. We are now seeing escalations in the Ukraine conflict with the Ukrainian attacks on Russia's bomber fleet, this is concurrent with the peace negotiations.
Some people have claimed of course that Trump is "Nixon 2.0" as there are certain commonalities in their policies : perhaps as a student of history this is something that might interest you.
Your thesis of decline regarding a decline of science may not exactly be true, if for example the US dominance in AI can allow for accelerated scientific progress.
10 comments:
Doesn't any AI dépend on it's sources?
Is this article serious?
There is no danger of AI. We need to convert the NNSA labs into AI labs and replace those scientists. AI can do everything.
We've reached the point where life imitates art:
Open the pod bay doors, HAL
I'm sorry Dave. I'm afraid I can't do that.
The first AI I was able to check by chance came mainly from an article I had seen the day before. There were three numbers in the paragraph, say a, b, c. b was four times a, c was four, and it was said that b=c×a with c=4. But the AI got it all wrong. It changed b so it was no longer 4a, kept a unchanged, but still said that b was 4 times a. Stupid. Recently I searched for an update on the Pike County, Ohio, murder cases. The AI search said the relevant county was in Missouri, said the trials had concluded when in reality one has not started, got two of the involved correct but did not mention the other two people including the pending trial, and one or two other mistakes, all in one not at all long paragraph. I have seen recently that for AI generated paragraphs there is often a disclaimer that there could be mistakes. DAH.
For several years I have used Google Scholar to find citations of papers. Recently the search was switched to AI. One of my papers was listed by AI with over 800 more hits than before. I would take this but it is incorrect and I have checked. I have recently read that AI for Google is very much slower than the previous methods. If you are using AI, check.
5/26/2025 4:34 PM
LANL sponsored a AI Jam session at a casino and they got close to a thousand people to ask Open AI questions and help it learn. The general sense was total disappoint on the part of the participant but Open AI got some good information for free. Many people said you simply cannot trust the AI for anything remotely complex. It also can screw up a lot of even even simple things when you had any kind of slight complexity to it. If you know the stuff you can find the issues and fix them. The real danger comes from people who do not know the stuff that well and will simply trust whatever the AI says as they have no way to check.
If just recognizing that three numbers should be related is considered as complex, then AI is pretty stupid and is useless except for trivia issues. Not at all surprised. Hype is ruling our world and bringing us down, not only in AI. When I retired, people asked me what I was going to do. I told them I was going to watch the decline of science at the Labs and the decline of the Western societies. The decline started somewhere around the Vietnam War era. It is happening, note drugs in our societies. And do not place the blame on the drug cartels. The drug situation is a classic case of suppy-demand. If there is demand, there will arise a supply. I did not know, and still do not, know the details of how things would play out. That is why I am watching. All is on track.
An example of hype that can be argued to mark the start of the decline of the US. The US government hyped that North Vietnam attached US ships in the Gulf of Tonkin. Congress used this hype to pass the Gulf of Tonkin Resolution, a thinly disguised declaration of war. Years later McNamara admitted in his memoir that the attach never happened.
A lot of this doomsday talk about LLMs being more dangerous then nukes or taking over the world is self serving. The companies are trying to make headlines and say, "Look at us, it's so powerful that it's dangerous!". Well, guess what, we could also hook a machine gun up to a computer and program it to shoot people. But we don't generally do that because it's known to be dangerous. Similarly, don't hook up an LLM to the Internet and let it have free reign. It's a machine, if it starts to malfunction, turn it off. Better yet, keep it in a sandbox. The ability to do stupid stuff with computer networks isn't exactly new.
6:19 -- I think it is good to understand history. But there were other wars started on false or inflated pretexts, for example the sinking of the Maine in Havana is now thought to be accidental, it was hyped to create the Spanish-American war.
It is interesting of course, that the Vietnam peace deal was preceded by the large Linebacker and Linebacker II attacks on Hanoi, to coerce the Vietnamese to the bargaining table. We are now seeing escalations in the Ukraine conflict with the Ukrainian attacks on Russia's bomber fleet, this is concurrent with the peace negotiations.
Some people have claimed of course that Trump is "Nixon 2.0" as there are certain commonalities in their policies : perhaps as a student of history this is something that might interest you.
Your thesis of decline regarding a decline of science may not exactly be true, if for example the US dominance in AI can allow for accelerated scientific progress.
Post a Comment