Vivaldi won't allow a machine to lie to you
-
@julien_picalausa said in Vivaldi won't allow a machine to lie to you:
@fratquintero: To the best of my knowledge, there is no solution that has the right guardrails. Can you point to a model that was trained using only data that is part of the public domain or obtained with consent and for which the pre-trained part was done without using of underpaid labor? Unless there is such a thing, all options we have are basically exploitative and cannot be condoned.
There are several solutions FOSS, eg. https://huggingface.co/allenai/OLMo-7B
-
@Catweazle The dataset used for that one claims to be "a diverse mix of web content, academic publications, code, books, and encyclopedic materials." No mention of whether it was all obtained with consent.
-
@julien_picalausa, well, there are some others "vanilla" in Hugging Faces, used to make all the examples for al kind of tasks you can find in HF Spaces. There you can find also al kind of models and datasets.
You may also contact the Andi Team because the LLM they use, they are the first in introduce AI in the search and with a philosophy very similar to Vivaldi, respect users and privacy. -
I miss the "Summarize this page" functionality Brave offers. I happily pay Brave $15/month to get Claude Instant via LeoAI. I'd happily do the same for Vivaldi's version of this.
-
@apt223 Won’t happen. Get some other AI tool and use it through an extension or as webpanel. It’s not like Vivaldi is stopping you from doing that. The tools are all there.
-
@luetage Yeah, but no: I value my time. I want something to work out-of-the-box or with minimal config, and I'm willing to pay for it. I'm not going to "linux my way" through a browser, especially when competing offerings have the functionality I'm looking for.
-
@apt223. as said before, I use Andisearch as main search and the Perplexity Extension, which offers me prvacy and fits my needs. But this, my needs, not necesarly those from others. I don't need an Chatbot from big corporations, nor that they log my use of those, above to need an paid account to use them, with which I give them personal data they can sell.
When I occasionally need other functions, I search for the corresponding services on Futurpedia, Future Tools or on Hugging Face, spaces, there are more than 7000 AI apps and services for every task I may need.
There is no need for Vivaldi, like others do, to put an AI at their convenience in the browser that may or may not be useful to me, apart from adding more work and resources to a small team. -
@julien_picalausa I agree this blog argument mostly but have one question. Chatbot by LLM lies and has copyright problem but summarization by LLM doesn't have such problems, does it?
If summarization also has a possibility to cause a copyright problem, translation also has a same problem, doesn't it? (This may be unnecessary worry because I don't know a detail of Lingvanex). -
@lamrongol, summarize or explain an article by LLM don't have copyright problems, also not an search engine with AI. Problems are others, an ChatAI, which answers are based on pretrained databases and not on the netcontent in real time, cn cause wrong answers generated when he don't have the answers based on reliable sources, image or video generators, which results don't have copyright (a bot can't reclaim copyright for it's creations, only the app as such can have copyright) but the risk is that videos and images can serve eg. for fake news, with serious political problems and consequences.
Generally copyright is a minor problem in AI apps and services, but missinformation, serious privacy problems in huge part of these and malicious use. -
@lamrongol LLMs can be decent at summarization, but there is a constant risk that they may introduce bias by choosing which part of the content to prioritize, or simply they may omit parts that are relevant to the reader. This is something I have experienced first-hand.
However, even if LLMs were amazing at summarization, the fact that the major players are exploiting all published text without permission and exploiting cheap labor in addition makes should give you pause. It is unclear to me whether a smaller model trained on an ethically obtained data set would be as good at the task.
At any rate, the work generated as a result of summarization is much less likely to violate any copyright, assuming you own the rights to the document you are asking summarization for. You may have to check whether the provider of the model is retaining your input for later training purposes, however.
When it comes to Lingvanex, I have been trying to get more details of what solution they are using. But regardless, it likely suffers from biases and mistranslations can occur, as has usually been common with automated translation. Again, copyright should not be an issue, so long as you own the rights to the document you are asking translation for.
If you need a translation for an official purpose however or otherwise cannot risk miscommunication due to a translation error, paying someone to do the translation for you will almost always guarantee a reliable result and will give you someone to keep accountable if the resut is unsatisfactory anyway.
-
@julien_picalausa, some days ago I read an article in Nature and I tried to summarize it using the Perplexity extension without result waiting several minutes (rotating donut), after this I searced this article with Andisearch, in the search results first summarizing it in 2 seconds with reliable result, after this requesting an explanation, in few seconds the correct result.
Yes, there are differences. IMHO Andisearch is very worthy in any aspect of being on the list of search engines, it's reliable, private, friendly devs and does a good job, without great Chat functions, it has but this isn't it's proposit. -
Generative AI is a plague that is showing up everywhere I turn. I'm glad I have finally found a browser that eschews it. I have swapped to Vivaldi primarily for this reason after being a long-time Opera user and am very satisfied so far.
-
@DavidSSabb94 welcome aboard!
-
@DavidSSabb94, welcome and enjoy. Yes, the generative Chatbots are a plague. AI can be be usefull in certain tasks, the above mencioned Andisearch is a good example and until now the only AI I found which is worth to use it daily.
It's devs have a very similar philosophy to the one of Vivaldis respect of what a AI shall do and what not and in privacy matters.
It can be added as any other search engine, no need to follow the hype to inbuild an stock chatbot in the browser, which don't make sense, it's way better that everyone, if they want, use an specific AI to their real needs, selecting from the thosands of existing AI for every task. -
@Catweazle I have found precisely one use for generative AI so far - when I was stuck trying to find evidence to back up my definition of a technical term. In the old days you could search two similar definitions and see which came back with the most hits... for some reason it just doesn't seem to work like that any more. In fact, I couldn't get any proper answer from a number of search engines, so I turned to the corporately-firewalled ChatGPT search instance (yes, my company is one of those betting all-in on AI).
I was pessimistic, but noticed with some amusement that the definition it came back with was very similar to mine and it certainly seemed to rule out the alternative definition that my colleague was backing. I'm aware how it works, so was careful to pose the question correctly and interpret the answer knowing it was based on commonly-connected phrases found on the internet, but in this one specific case it did seem to provide an extra bit of confidence.
Having said that, I would avoid using it for any normal searches.
-
śmiech na sali ,ta wy nie potraficie zrobić dobrej synchronizacji , zakładek ,mini ikonek, ustawień, dodatków i rozszerzeń ,na pocieszenie chrome zjebał to na waterfox, firefox i u Was,co mnie bardzo boli zakładki mam szare a oczy stare a zakłądek 17 tyś do pracy....
-
Do you think AI will make humans lazier and defeat human intelligence?
-
@manueins said in Vivaldi won't allow a machine to lie to you:
AI will make humans lazier
Yes, i fear that lazyness will lead to people who are forgetting what learning means.
-
To use AI in first line is needed HI (Human Intelligence). It's a big difference to use AI to summarize or explain a long given essay, which certainly can help in the studies and which LLMs can do really good, and other is to simply ask an ChatGPT, copy-pasting the result as is, without checking it.
-