I have just spent an interesting hour on the PC since my last post. Those who read it will recall that I posted some modern Greek, and then the Google Translate output for it – good, but by no means perfect. It then occurred to me to try Microsoft’s Bing AI. The output from that was marvellous, so I added that to the end of the post.
Then I started pasting more of the modern Greek into Bing AI, chapter by chapter. This went well until chapter 11, when it crashed. And crashed and crashed. When it did return, the output was obviously inferior, using different sources! So I stopped. The Bing AI interface was frankly a mess anyway.
Then it occurred to me to try Bard AI, the Google product. This produced… inferior output. Not really any better than Google Translate.
Then I tried ChatGPT 3.5, the OpenAI product. And suddenly I was getting good output again, if subtly different.
Google has dominated the web ever since the late 90s. But it is basically a search engine company. It rose to power because it was a markedly better search engine than AltaVista, which it destroyed. But the founders of Google have long since departed, as has its “don’t be evil” ethos. For many years now the company has concentrated on squeezing dollars out of the web.
But “AI” is basically search-engine technology. So the rise of OpenAI is really the arrival of a new generation of search engine; and Google is struggling.
What the OpenAI people have realised is that most people do not, actually, want to “search the web”. They want information. So they have reimagined the search engine, added a layer of algorithms on top, with a chat-bot on top of that. Their search engine produces information, not lists of results. These engines can be targeted for developers, or translators, or other specialised markets.
It looks very much as if OpenAI have created the next generation search engine.
If so, it must have become obvious to search-engine people about a year ago. Google executives must have been quaking in their well-funded boots. This is the end for Google, as Google was the end for AltaVista… unless. Unless Google can catch up.
It has been telling how little we really know about how “AI” works. The hype has been tremendous. The hard facts have been few. And no wonder, if it’s now a cut-throat commercial race.
Google may have some trouble here. It isn’t really a tech company any more. It’s an advertising company. Products like Google Docs were invented elsewhere and bought in. Microsoft went down this route, as IBM did before them. There is still interesting tech going on, but they aren’t the cutting edge.
Microsoft will be the big beneficiary of this new era, because of its far-sighted investment in the OpenAI company. Bing AI, I think, must basically be a front-end on OpenAI.
If all this is true, then Google may not be able to fight back. The link to Microsoft means that they cannot just buy OpenAI, which would have been a possible way forward. So… maybe Google will enter the long decline experienced by other tech companies before. Once people do not use their search engine, it’s basically over.
For the conspiracy minded, this may also explain some otherwise perplexing events of the last year. In particular it may explain the boardroom coup at OpenAI, when the directors sacked the company founder, Sam Altman. Why the heck would you do that? It didn’t stick – Microsoft intervened. The directors seem to be just corporate drones, of no special talent. Microsoft wasn’t going to let them wreck the company that was about to take over the world. But why would you do that at all? Why get rid of the guy who will make you rich?
If this was a Hollywood movie, who would be the prime beneficiary of those mysterious shenanigens at OpenAI? If OpenAI had collapsed, if Altman could no longer direct the company, who would benefit? It’s an interesting thought. Did someone bribe the renta-board of OpenAI to crash the company? Some existing vested interest? Not necessarily Google, of course, but possibly someone with shares in it?
Likewise it is interesting to see all the lawsuits, and claims of copyright on the databases used, the large language models (LLMs). Not that the claims are bogus. If I understand correctly, there is little doubt that many of the modern AI databases are indeed based upon masses of copyright material. We know this thanks to one company making an incautious publication which listed what they were loading into their database!
But say that the copyright claims probably valid. Whether that is true or not – and when it will be decided – will be a decision for the US courts. These are notoriously political, so money and power will no doubt decide. Again, who would benefit from knobbling the new technology?
In truth I have no idea about all of this. There must be people less connected than myself, but it might be hard to name them! But perhaps all this really signals that the revolution is here.
Yet this revolution may not benefit us, the ordinary internet users.
For one thing, OpenAI is the first web technology that you have to pay for. It’s on subscription. This is a new, and unwelcome change to the free internet.
Likewise searching the web at least gives you a choice of viewpoints. If you get your news from OpenAI, you get one viewpoint only. It’s great news for the would-be censors, who seem to grow more numerous every day.
It will be interesting to see what happens.
I ran the same passage through the free version of DeepL with the result below, which has a few improvements but with the strange “a-bend” and “eighths” for half:
3. Nicholas came from Patara, a town in the province of Lycia, which at that time (of Saint Nicholas) had many inhabitants. But now it looks more like a comune than a town14. And this is what happens to cities in many ways-they fall into decay and their inhabitants emigrate-because of their innumerable judgments against God. Thus we also understand what we are wont to write, namely, that cities are punished for the sins of their inhabitants15. That is, there is a lowland place nearby, torn in eighths, like a torn garment. And from the rift that has been created, hot steam seems to rise by day, while at night fiery smoke rises like from a bronze furnace16. And it burns the hand of whoever dares to a-bend it, but without devouring it or the flesh, as would be the case if one were to stretch out his hand into the fire; but it is an illustration and prelude, with little resemblance, to the eternal fire. And being in wonder at the phenomenon, the elders in old age conveyed from generation to generation, each father to his child, (as this unwritten but true saying is preserved to us), that because of the licentiousness and carnality of its inhabitants, divine judgment condemned this land to this horrible sight, to prevent their return to this place.
Translated with DeepL.com (free version)
How did you use OpenAI to get satisfactory results?
I’m (hesitantly) excited about (some) possibilities of AI. Yet I also balance out my fears generated by, say, the movies “The Terminator” (1984), it’s first sequel (1991), and Kubrick’s spacefaring masterpiece (1968). The reality is that it most likely will remain under human control and not run amuck — as money will, like everything else, be the controlling agent.
I recently gained a few AI certifications and built a history-specific app on the CGPTStore, and will continuously craft it for accurate, reliable output. This is necessary since there is SO much unreliable information online which can creep into the output.
I appreciate your detailed writing about this, and within the next few days will run some of my ancient sources through and see if our results match. And there are free tiers of OpenAI, and it’s not the only game in town. Right now, for me, $20/mo is a tiny price to pay for being with some of the emerging ground floor innovations, stealthily surfing the rising wave, while I lament on missing out on so many tides before.
OpenAI is one-size-fits-all, take it or leave it; and you’ll pay for the “privilege”.
In Greek we split words with a -. The word that throws you off is split. Also as mention 8υο should be δύο, meaning two.
That being said AI which currently has a hype is simply a better method to do classification. If you read any of the news item on how AI found new galaxies in some survey, read it as one of the new non-parametric classification methods such as artificial neural networks or gainful adversarial networks classified real data (as opposed to synthetic) better that more traditional parametric classification methods such as maximum likelihood or K nearest neighbor and pull out more data. The whole Large Language Model buzz is about a new classification method that can pull out of vast data better matches. Better non parametric classifications that can better fit the thin distinctions that human based categories have that in physical space are not big but in conceptual space are. Then again this is how I understand it as a remote sensing scientist.
@Phil Turner, I went to https://chat.openai.com/, signed up for a free account. Then I put in the box “can you translate something from modern Greek to English?” It said yes and asked me to paste in the item; then I popped the text from the preceding blog post in. I repeated the process to do more chapters. There’s probably a better way, but that was what I did. I did the same at Bard and Bing.
@ikokki: Thank you for picking that up! I had a bunch of soft line-breaks in there from Word as well. I’ve fixed those and updated the previous post. I also added in the ChatGPT output. Curiously the BingAI is still not working properly, and just gives stuff from Bing Translate.
@Paul Chandler: I hadn’t heard of DeepL! I must give it a go. Sorry about my rubbish OCR. I’ve fixed it – have another go!
@Paul R Pearson: I agree, in the tech world this is now becoming something that you have to know about to get a job. I know almost nothing. It was just an idle thought to try it as a translator.
I find the web interface cumbersome. Is there a command-line interface?
I tried DeepL and got this. I think DeepL is a pre-AI translator, like Google Translate.
Nicholas came from Patara, a town in the province of Lycia, which at that time (of Saint Nicholas) had many inhabitants. But now it looks more like a comma than a town14. And this is what happens to cities in many ways-they fall into decay and their inhabitants emigrate-because of their innumerable judgments against God. Thus we also understand what we are wont to write, namely, that cities are punished for the sins of their inhabitants15. That is, there is a lowland place nearby, torn in two, like a torn garment. And from the rift that has been created, hot steam seems to rise, by day, while at night fiery smoke rises like from a bronze furnace16. And it burns the hand of whoever dares to a-bend it, but without devouring it or the flesh, as would be the case if one were to stretch out his hand into the fire; but it is an illustration and prelude, with little resemblance, to the eternal fire. And being in wonder at the phenomenon, the elders in old age conveyed from generation to generation, each father to his child, (as this unwritten but true saying is preserved to us), that because of the licentiousness and carnality of its inhabitants, divine judgment condemned this land to this horrible sight, to prevent their return to this place.
Translated with DeepL.com (free version)