#chatgpt
After a little more playing around with GPT4-o, I think it's added some fun new features, but isn't significantly better at anything that really matters compared to its previous incarnations. It's also restricted for free users to an absurd and almost offensive degree : it's barely enough to tell if it would be worth paying for, let alone do anything useful with.
Some tests : I fed it links to five arXiv papers and asked it to identify the first author. It got four of them wrong, and managed to correct one more when I told it so. Interestingly, when I fed it screenshots of the same papers, it identified all the authors correctly on the first time. I deliberately cropped the screenshots differently to show different parts of the text, sometimes truncating the title/abstract and sometimes not. This didn't seem to make any difference.
When I fed it a whole PDF, it could also identify the first author correctly, but as with other LLMs capable of handling PDFs, its results to simple inquiries were a uselessly mixed bag. It got a numerical estimate from a figure completely wrong the first time, but corrected it to something far more sensible the second time (I didn't ask it to correct the value or suggest a better one, I just asked for an explanation of how it got the number - ChatGPT still insists on treating any question about its answer as a request for a correction). On another figure it just came up with a nonsense value that bore no relation to anything in the figure at all.
So nothing has really changed here. Advertising this is "free" when you get ~10 messages and maybe ~5 files every 24 hours is like Amazon's continued insistence on giving me two free months of Audible : optimistic at best. Multimodal features are nice to have but near-useless without having fixed the hallucination problems, like giving games better and better graphics but forgetting to fix bugs in the gameplay. I know, it's not the same, and hallucinations are a fundamental part of how LLMs work. But still, continuing to add ever-more and more sophisticated ways of accessing inaccurate information feels very much like a house of cards. Yes, these things are still good for inspiration, and there are many cases where accuracy isn't needed. Still I would have hoped for a bit of progress here, but there hasn't been any as far as I can tell. That's worrying.
So, I have access to ChatGPT-4o now. So far, so... mixed.
Q : Could you tell me a little bit about this paper: https://arxiv.org/abs/2312.03601
[This is the one I posted yesterday]
A : Generic though not inaccurate response, except for : "The authors, led by Jonathan H. C. Rhee..."
This is not one of the authors. The first author is Xiao Li. Let's skip the boring and unhelpful discussion as to why it came up with this and get to the science.
Q : Can you spot any inaccuracies in the paper?
A : Identifying inaccuracies in a scientific paper requires a thorough understanding of the subject matter and often involves detailed analysis and peer review. If you have specific aspects of the paper you want to investigate for potential inaccuracies, I can help by examining those areas closely or by summarizing relevant findings from the text.
[Oh, thank you so much, how helpful !]
Q : Well that's why I'm asking you. Consider the major message of the paper and see if there's anything they present that contradicts this. Pay special attention to the appendix.
A : Long-winded generic answer about all the usual concerns about sample selection, data quality, statistics, etc., the kind of stuff that's true about literally every paper, e.g. " The authors could have provided more detailed explanations or additional analyses to support their conclusions, particularly when claiming the uniqueness of certain galaxy properties."
[Well, duh.]
Q : I'm asking if the paper makes any specific claims which are contradicted by other statements they make. I don't mean that their data might not be of the best quality or their methods flawed, I mean specific, direct contradictions.
A : Upon closely examining the paper "On the existence, rareness and uniqueness of quenched Hi-rich galaxies in the local Universe" and its appendix, one specific contradiction stands out:
Main Claim:
The paper's major claim is that they have identified a sample of quenched (i.e., no active star formation) yet HI-rich galaxies (referred to as RR galaxies) and that these galaxies are rare and unique in the local universe. The study emphasizes the rareness of such galaxies by comparing their HI richness to that of typical quenched galaxies, which are usually HI-poor.
Contradictory Statement in the Appendix:
In Appendix A, where the authors present optical images of the selected RR galaxies, they show the HI-to-stellar mass ratios for these galaxies. The paper discusses these galaxies as being "red but HI-rich." However, upon closer inspection of these images and the related text, it becomes apparent that some of the galaxies display features that are inconsistent with a fully quenched status. Specifically, some images suggest the presence of spiral arms or other structures typically associated with ongoing star formation, which contradicts the classification of these galaxies as fully quenched.
Bam. Perfect. Especially because the authors themselves do not remark on this. The damn thing can't read a bloody author list but it can spot features inconsistent with quenched galactic star formation. Thoroughly odd !
Ученые создали первый искусственный биологический синапс
Команда физиков из Утрехтского университета в Нидерландах и Университета Соганга в Южной Корее создала искусственный синапс — биологическую основу способности человеческого мозга мыслить и рассуждать. Ранее было предпринято множество попыток воспроизвести способ, которым наш мозг обрабатывает сложную информацию, но обычно они полагались на кремний и металл. Однако новый синапс, разработанный исследователями, использует обычную воду и …
Сообщение Ученые создали первый искусственный биологический синапс появились сначала на AB-NEWS.
#нейробиология #chatgpt #мозг #нейроны #lang_ru #ru #abnewsru #abnews #наукаитехника
Ian Stewart ist einer der populärsten englischsprachigen Mathematiker. Sein Buch veranschaulicht, wie stark unser Leben von angewandter Mathematik geprägt ist. Eine Rezension
Wer verstehen möchte, wie zahlreich und wichtig Anwendungen von Mathematik für unseren Alltag sind, wird in diesem spannenden Buch fündig. Eine Rezension (Rezension zu Die Welt als Zahl von Ian Stewart)#Mathematik #Technik #KünstlicheIntelligenz #KI #Naturwissenschaft #Einstein #Relativitätstheorie #Quantenphysik #Algorithmen #Digitalisierung #ChatGPT #Astronomie #ITTech #Physik
»Die Welt als Zahl«: Ob Navi oder Organtransplantation: die Präsenz der Mathematik
#LLM Agents can Autonomously #Exploit One-day Vulnerabilities
Source: https://arxiv.org/abs/2404.08144
To show this, we collected a dataset of 15 one-day vulnerabilities that include ones categorized as critical severity in the #CVE description. When given the CVE description, GPT-4 is capable of exploiting 87% of these vulnerabilities compared to 0% for every other model we test (GPT-3.5, open-source LLMs) and open-source vulnerability scanners (ZAP and #Metasploit).
#ai #technology #Software #chatgpt #bug #hack #news #cybersecurity
Christoph Drösser fragt sich, welche Auswirkungen die zunehmende Verbreitung von KI-Anwendungen auf unsere Sprache hat. Seine Antworten sind klug und differenziert. Eine Rezension
Die KI wird immer alltäglicher. Aber was bedeutet das für uns und unsere Kommunikation? Dieses gelungene Büchlein gibt erste kluge Antworten. Eine Rezension (Rezension zu Was macht KI mit unserer Sprache? von Christoph Drösser)#KI #künstlicheIntelligenz #ChatGPT #Bot #Chatbot #Sprache #Sprachkompetenz #Lesen #Schreiben #Prompt #Turing #Kultur #ITTech #PsychologieHirnforschung
»Was macht KI mit unserer Sprache?«: Und täglich grüßt ChatGPT
Is #ChatGPT corrupting peer review? Telltale words hint at #AI use
Source: https://www.nature.com/articles/d41586-024-01051-2
Out of more than 1,600 scientists who responded to a 2023 Nature survey, nearly 30% said they had used generative AI to write papers and around 15% said they had used it for their own literature reviews and to write grant applications.
Модель искусственного интеллекта Apple «превосходит GPT4», говорят исследователи
Apple продолжает работать над своим ИИ-помощником на устройстве, и, по мнению ее собственных исследователей, они уже готовы опередить крупнейшего игрока в отрасли. В статье, опубликованной на сайте препринтов arXiv, специалисты Apple AI похвастались, что их локальная модель «существенно превосходит» GPT4, технологию, лежащую в основе ChatGPT, Google Gemini и Microsoft Copilot. Во многом это связано с …
Сообщение Модель искусственного интеллекта Apple «превосходит GPT4», говорят исследователи появились сначала на AB-NEWS.
#компьютерыинейросети #chatgpt #информация #искусственныйинтеллект #lang_ru #ru #abnewsru #abnews #наукаитехника
