NLA存在幻觉问题，可能会生成与实际内容不符的虚假信息，这是其重要局限性之一。

NLA存在幻觉问题，可能会生成与实际情况不符的描述，这表明当前技术仍有局限性，需要结合其他验证方法来确保解释的准确性。

这一局限性揭示了当前AI可解释性技术的挑战，提醒我们在解读NLAs结果时需要谨慎验证，不能完全依赖其单方面描述。

🤖【令人震惊的数字】幻觉率降低 52.5%——这是 OpenAI 有史以来在单次模型更新中宣称的最大幻觉降幅。更重要的是这发生在医疗、法律等高风险领域。幻觉是 AI 在专业服务场景落地的最大障碍，这个数字若属实，意味着企业 AI 可信度的拐点正在到来。

A third of respondents admit they've seen 'at least once' negative consequences of inaccurate output. That sounds low, as 100% will have been given hallucinations. So 1-in-3 doesn't catch them all before they run-up damage. (vgl Deloitte's work in Australia)

If you like the annotated conversation here, you can also join in on the margins of the paper itself.

In our paper, we address this idea that ALL ✨sparkling intelligence✨ outputs are generated using the same technology and practices. We argue that it is useful to have a term for those outputs that don't match our shared reality or factual requirements, and for that we propose "mirage".

In my blog post introducing our paper, I suggest AI "rainbows" as a term for mirages that we do value.

I'm assuming this is some kind of ✨sparkling intelligence✨ and given that Dr. Oblivion seems to miss the point of the paper and our discussion here, I found it more illustrative than helpful ;)

Agree! This is a point we make in the paper. And then there are those mirages that we actually value, which I have proposed the term "AI rainbows" in my blog post introducing our paper.

Turns out the story is a bit more complicated than that, at least according to the history shared by another participant below.

This is a great history of the term "hallucination" in the discourse of ✨sparkling intelligence✨ — huge thanks to whoever shared it! I've also added it to our collaborative bibliography.

"lemoine" has succumbed to the superficial appearance of the conversational sequence. The LMM has recognition of patterns of English conversation and knows the sorts of patterns that could come next in the word sequence. This is without reference to any understanding. The lack of understanding is why the misnomer "hallucination" is inappropriately used for some patterns like footnotes that correctly follow the footnote pattern but don't actually refer to an existing source.

You might also want to visit my blog post, where I introduce the publication of this paper alongside some additional ideas on interventions to prevent AI mirages, on AI mirages vs AI rainbows, and on how AI terminology plays out in different disciplines.

Diaspora: the jews living outside Israel (https://www.merriam-webster.com/dictionary/diaspora)

Interesting musical roots for Reggae... Wonder if this is still present?

Mystical roots.

(Note, I give this the fiction tag because I might want to look into this mystical religion for fiction writing as inspiration)

Logical that marijuana (a drug) is correlated with the mystical concept of communicating with diseased spirits for marijuana makes you hallucinate (or perhaps it's demonic in nature?)

He acts as an echo, and an echo is simply just a reaffirmation, a suspicion being ingrained, a hallucination

https://claude.ai/onboarding

"Reading with an artificial intelligence" seems like a fascinating way to participate in the Great Conversation.

Wordcraft's hallucinations can create interesting and creatively surreal suggestions.

How might one dial up or down the ability to hallucinate or create surrealism within an artificial intelligence used for thinking, writing, etc.?

Furl, N., McKay, R., & Coltheart, M. (2022, January 29). The Paradox of Delusions: Are Deluded Individuals Resistant to Evidence?. https://doi.org/10.31234/osf.io/e37c2

This is a very crucial topic. Fear stands among the leaders of bad decisions' motivators so when you'll grasp the depth of meaning for this subject your life will never be the same.

Let us consider briefly what reasons make you scared. First and foremost you must be thinking that this event or person is absolutely real. The follow up is the idea: this situation threatens you somehow. And final step to get you frightened is to assure you that you have no control.

The combo of these reasons leads you to conclusion you might become a victim so you need react preventively right now. This is a very nasty hook which you can dodge by realizing: all of those statements are equally untrue.

Take time to learn what's in the quotes related, without this solid foundation forgiveness can't be understood.

The correction of fear is your responsibility. When you ask for release from fear, you are implying that it is not. You should ask, instead, for help in the conditions that have brought the fear about. T-2.6.4

God did not create a meaningless world. W-14

I am not the victim of the world I see. W-31

I have invented the world I see. W-32

The world you see is an illusion of a world. God did not create it, for what He creates must be eternal as Himself. Yet there is nothing in the world you see that will endure forever. C-4.1

Forgive yourself the thought He wanted this for you. W-99.7

What if you recognized this world is an hallucination? What if you really understood YOU made it up? T-20.8.7

The end of dreaming is the end of fear T-28.3.4

If I defend myself I am attacked. W-135

How safe the world will look to me when I can see it! It will not look anything like what I imagine I see now. Everyone and everything I see will lean toward me to bless me. I will recognize in everyone my dearest Friend. What could there be to fear in a world that I have forgiven, and that has forgiven me? W-60.3

I thank You, Father, for Your plan to save me from the hell I made. It is not real. And You have given me the means to prove its unreality to me. The key is in my hand, and I have reached the door beyond which lies the end of dreams. W-342.1

Renault, M. (2020, May 5). ‘ICU Delirium’ Is Leaving COVID-19 Patients Scared and Confused. The Atlantic. https://www.theatlantic.com/science/archive/2020/05/coronavirus-icu-delirium/610546/