I’ve been working on my Theory of Everything physics framework and I’d advise a lot of caution when using Claude and its competitors.
The main book (out of 20-book non-fiction and fiction portfolio) has about 100K words, half of it mathematical derivations, spread out across 450 pages.
The whole premise of the framework is that the universe is discrete and deterministic at its most basic structure, so I used well established Finite Element Analysis to calculate the known measurements of particle physics with great success.
So far, that’s great as the computations check out. But the problem is with editing these books.
I have a project folder with hundreds of files with research notes, programs, guides with protocols for due diligence, as any reputable science or engineering research project would require.
Nevertheless, even Claude Opus 4.8 and Fable 5 at their highest setting have made glaring mistakes and confabulations that stray from the canonized conclusions and calculated results from the project.
My other side-project is a neuromorphic AI that does not hallucinate like LLMs are prone to do. I use it as a fact checker and it caught many mistakes and discrepancies even after Fable spawned up to 75 agents and spent close to 100 million tokens reviewing a book with 100K words.
The moral of the story is that LLMs are not to trusted. Be very judicious when using these transformer-based models. They confabulate. And that’s why I built my own AI that is founded on a different type of neural network which does not make shit up. It abstains or makes a suggestion, but it does not generates junk like Claude does.
As the Dark Lord has said, Ai is a force multiplier for the good, true and beautiful. Can’t remember if he added the GTB. I refrained from parentheses. Twice.
I translated a friend's book from German to English last year, even though I only know a smattering of German. I used all three AIs to check each other: ChaptGPT, Claude, and Grok. It only took two hour-long telecons with the author to decide on a few terminological translations for some of the specific terms of art he included in the book. By the end of the process, after asking the AIs to recreate the prompt, it ended up about four typed pages long. The original book was very technical and wordy, in other words, very German. My translation was much looser and more readable, as needed for a non-philosophical English-speaking audience. Tuning the AIs to do what I wanted: to keep the ideas and semantic content, but use simpler words and sentences, was very difficult. A strict transliteration would preserve the author’s words but would not make for a readable book. The point is to communicate the ideas, not preserve the words. But I get your point.
This is interesting, but would like if someone that reads those languages could chime in. My recent experience, a few weeks ago, with some of the LLMs was not the best when it came to translating technical texts from one modern language to the other and had to redo the document partially. But I admit I have to practice with orchestration (having one agent verifying the output of other agents and so on).
I'm thinking of doing this for Greek, too. Opus 4.5/4.6 did a bang up job of translating Job and Esther from the original Greek, actually more readable than any existing translation I've read.
For whatever reason, Claude refused to do it, but API calls were more than able to handle it.
This is the stuff I had hoped you pointed your translation skills at. While I prob wont get to read all of these and much of it i may not be capable of understanding this work seems so incredibly important. thank you Vox and Scott for doing this.
I’ve been working on my Theory of Everything physics framework and I’d advise a lot of caution when using Claude and its competitors.
The main book (out of 20-book non-fiction and fiction portfolio) has about 100K words, half of it mathematical derivations, spread out across 450 pages.
The whole premise of the framework is that the universe is discrete and deterministic at its most basic structure, so I used well established Finite Element Analysis to calculate the known measurements of particle physics with great success.
So far, that’s great as the computations check out. But the problem is with editing these books.
I have a project folder with hundreds of files with research notes, programs, guides with protocols for due diligence, as any reputable science or engineering research project would require.
Nevertheless, even Claude Opus 4.8 and Fable 5 at their highest setting have made glaring mistakes and confabulations that stray from the canonized conclusions and calculated results from the project.
My other side-project is a neuromorphic AI that does not hallucinate like LLMs are prone to do. I use it as a fact checker and it caught many mistakes and discrepancies even after Fable spawned up to 75 agents and spent close to 100 million tokens reviewing a book with 100K words.
The moral of the story is that LLMs are not to trusted. Be very judicious when using these transformer-based models. They confabulate. And that’s why I built my own AI that is founded on a different type of neural network which does not make shit up. It abstains or makes a suggestion, but it does not generates junk like Claude does.
Thank you Vox for posting this. Let a thousand translation pipelines bloom!
As the Dark Lord has said, Ai is a force multiplier for the good, true and beautiful. Can’t remember if he added the GTB. I refrained from parentheses. Twice.
I translated a friend's book from German to English last year, even though I only know a smattering of German. I used all three AIs to check each other: ChaptGPT, Claude, and Grok. It only took two hour-long telecons with the author to decide on a few terminological translations for some of the specific terms of art he included in the book. By the end of the process, after asking the AIs to recreate the prompt, it ended up about four typed pages long. The original book was very technical and wordy, in other words, very German. My translation was much looser and more readable, as needed for a non-philosophical English-speaking audience. Tuning the AIs to do what I wanted: to keep the ideas and semantic content, but use simpler words and sentences, was very difficult. A strict transliteration would preserve the author’s words but would not make for a readable book. The point is to communicate the ideas, not preserve the words. But I get your point.
This is interesting, but would like if someone that reads those languages could chime in. My recent experience, a few weeks ago, with some of the LLMs was not the best when it came to translating technical texts from one modern language to the other and had to redo the document partially. But I admit I have to practice with orchestration (having one agent verifying the output of other agents and so on).
This is really cool. I have a couple of old books that I have wanted to read in a more modern vernacular and this would work for that too.
I'm thinking of doing this for Greek, too. Opus 4.5/4.6 did a bang up job of translating Job and Esther from the original Greek, actually more readable than any existing translation I've read.
For whatever reason, Claude refused to do it, but API calls were more than able to handle it.
That makes a lot of history available, thanks.
Will you eventually branch out from old science books?
To a degree.
I've translated a number of Jewish works that gentiles might label as philosophical.
There are so many thousands. It's like the wild west.
This is the stuff I had hoped you pointed your translation skills at. While I prob wont get to read all of these and much of it i may not be capable of understanding this work seems so incredibly important. thank you Vox and Scott for doing this.
Thank you. If not Vox, then someone else!
Same, this is what i had hoped for as well, and possibly eventually Russian works as well.
That makes sense. They have a rich history.
Russian history is pretty much a closed book to the West.