3/5
Deux algorithmes universels en découlent naturellement :
• Choix des actions
• Auto-correction « selon les fruits » (apprentissage par l’échec réel et ajustement sensoriel)
4/5
Article complet (en anglais) :
https://medium.com/@leg.sorn/limitations-are-not-a-weakness-3cbb1dbb87b0
5/5
Co-écrit avec Grok (xAI)
#AGI #IntelligenceArtificielle #EmbodiedAI #Philosophie #IA #Limitations
Limitations Are Not a Weakness.

Limitations Are Not a Weakness. They Are the Source of Meaning How constraints create sense, living behavior, and the foundation for true artificial intelligence We usually think of limitations as …

Medium

Hallucination Is Inevitable: An Innate Limitation of Large Language Models

https://arxiv.org/abs/2401.11817

#HackerNews #hallucination #languagemodels #AIresearch #technology #limitations

Hallucination is Inevitable: An Innate Limitation of Large Language Models

Hallucination has been widely recognized to be a significant drawback for large language models (LLMs). There have been many works that attempt to reduce the extent of hallucination. These efforts have mostly been empirical so far, which cannot answer the fundamental question whether it can be completely eliminated. In this paper, we formalize the problem and show that it is impossible to eliminate hallucination in LLMs. Specifically, we define a formal world where hallucination is defined as inconsistencies between a computable LLM and a computable ground truth function. By employing results from learning theory, we show that LLMs cannot learn all the computable functions and will therefore inevitably hallucinate if used as general problem solvers. Since the formal world is a part of the real world which is much more complicated, hallucinations are also inevitable for real world LLMs. Furthermore, for real world LLMs constrained by provable time complexity, we describe the hallucination-prone tasks and empirically validate our claims. Finally, using the formal world framework, we discuss the possible mechanisms and efficacies of existing hallucination mitigators as well as the practical implications on the safe deployment of LLMs.

arXiv.org
Things C++26 define_static_array can’t do

We’ve seen previously that it’s not possible to create a constexpr global variable of container type, when that container holds a pointer to a heap allocation. It’s fine to create a global constexpr std::array, or even a std::string that uses only its SSO buffer; but you can’t create a global constexpr std::vector or std::list (unless it’s empty) because it would have to hold a pointer to a heap allocation.

Arthur O’Dwyer

Ukraine’s killer robots | About That
Five years into the war with Russia, Ukraine faces persistent manpower problems. Andrew Chang explores the advantages and limitations of using unmanned ground vehicles (UGVs) on the front line, in Ukraine and beyond.

Images provided by The Canadian Press, Reuters, Adobe Stock and Getty Images
https://www.cbc.ca/player/play/9.7176908?cmp=rss

Square dances and cell signal dead zones: Rangers confront Canada’s Arctic reality
A 52-day, 5,200-kilometre Canadian Ranger patrol along the Northwest Passage revealed the human and technological limits of Arctic operations. From impromptu square dances in remote communities to frozen batteries and fragile communications gear, the Canadian military is finding the North demands adaptation and partnership.
https://www.cbc.ca/news/politics/arctic-defence-rangers-patrol-9.7173700?cmp=rss
Square dances and cell signal dead zones: Rangers confront Canada’s Arctic reality
A 52-day, 5,200-kilometre Canadian Ranger patrol along the Northwest Passage revealed the human and technological limits of Arctic operations. From impromptu square dances in remote communities to frozen batteries and fragile communications gear, the Canadian military is finding the North demands adaptation and partnership.
https://www.cbc.ca/news/politics/arctic-defence-rangers-patrol-9.7173700?cmp=rss

They read themselves, or: As they are

A Sijo

some impose their inner template their flat idea, a label; they read me through their fixed frame seeing only their own beliefs; I have learned to accept them and hold them all as they are

Sijo?

A Korean verse form related to haiku and tanka and comprised of three lines of 14-16 syllables each, for a total of 44-46 syllables. Each line contains a pause near the middle, similar to a caesura, though the break need not be metrical. The first half of the line contains six to nine syllables; the second half should contain no fewer than five. Originally intended as songs, sijo can treat romantic, metaphysical, or spiritual themes. Whatever the subject, the first line introduces an idea or story, the second supplies a “turn,” and the third provides closure. Modern sijo are sometimes printed in six lines.

Let’s write poetry together!

When it comes to partnership, some humans can make their lives alone – it’s possible. But creatively, it’s more like painting: you can’t just use the same colours in every painting. It’s just not an option. You can’t take the same photograph every time and live with art forms with no differences.

Ben Harper (b. 1969)

Would you like to create poetry with me and have a completed poem of yours featured here at the Skeptic’s Kaddish? I am very excited to have launched the ‘Poetry Partners’ initiative and am looking forward to meeting and creating with you… Check it out!

#Acceptance #Assumptions #Confidence #Identity #Limitations #Poem #Poetry #Pride #SelfConfidence #Sijo

AI絵生成 Генерація зображень ШІ
AI絵生成ソフトで何処が失敗してるかと云えば、人間の体の構造や比率を充分学習させてないと云う点が大きい。
Основна помилка програмного забезпечення для створення зображень ШІ полягає в тому,

https://note.com/poison_raika/n/n501ac31af2f1

<>

#AI #3d #image #model #new #human #software #animals #traditional #layout #structure #generation #frequency #necessary #develop #vizualize #objects #dimensions #approach #methods #probably #limitations #adequately #irregularities #proportions

AI絵生成 Генерація зображень ШІ|ポイズン雷花

AI絵生成ソフトで何処が失敗してるかと云えば、人間の体の構造や比率を充分学習させてないと云う点が大きい。 各部位の配置や形状、詳しい構造等を3dモデルを絵に重ねて学習させる等の新しい仕法を開発する必要が有るのは明確です。 ただ画像を見せて学習させると云う従来の仕法では限界が有るのは明らか。特に手足の部分に乱れが非常に高い頻度で発生するのは皆さんもよく感じてるでしょう? 勿論、人間だけでは無い。動物や物においても充分学習出来てない部分が多々有ります。 後は見えない部分も又3dで可視化して、全体を3次元で解析出来るシステムの構築が必要でしょう。対象が3dか2dに関わらず、そう云うアプローチ

note(ノート)