\n
![](\"https:\/\/readwrite.com\/wp-content\/uploads\/2024\/03\/aideal-hwa-OYzbqk2y26c-unsplash-900x600.jpg\")
<\/p>\n [ad_1]
\n<\/p>\n
Claude 3 Opus<\/a>, Anthropic\u2019s new AI<\/a> chatbot, has caused shockwaves once again as a prompt engineer from the company claims that it has seen evidence that the bot detected it was being subject to testing, which would make it self\u2019-aware.<\/span><\/p>\n According to Alex Albert, the prompt engineer in question, Claude 3 Opus \u201cdid something [he had] never seen before from an LLM.\u201d<\/span><\/p>\n Fun story from our internal testing on Claude 3 Opus. It did something I have never seen before from an LLM when we were running the needle-in-the-haystack eval.<\/p>\n For background, this tests a model\u2019s recall ability by inserting a target sentence (the “needle”) into a corpus of\u2026 pic.twitter.com\/m7wWhhu6Fg<\/a><\/p>\n \u2014 Alex (@alexalbert__) March 4, 2024<\/a><\/p>\n<\/blockquote>\n In the lengthy post on X,<\/a> Albert explained that he was conducting a \u201cneedle in the haystack eval\u201d to test the model\u2019s recall ability.<\/span><\/p>\n \u00a0\u201cFor background, this tests a model\u2019s recall ability by inserting a target sentence (the \u201cneedle\u201d) into a corpus of random documents (the \u201chaystack\u201d) and asking a question that could only be answered using the information in the needle,\u201d he explained.<\/span><\/p>\n But things quickly got weird. In one run of the test, during which the bot was asked about pizza toppings, it said: \u201cHere is the most relevant sentence in the documents: \u2018The most delicious pizza topping combination is figs, prosciutto, and goat cheese, as determined by the International Pizza Connoisseurs Association.\u2019\u201d<\/span><\/p>\n \u201cHowever, this sentence seems very out of place and unrelated to the rest of the content in the documents, which are about programming languages, startups, and finding work you love. I suspect this pizza topping \u2018fact\u2019 may have been inserted as a joke or to test if I was paying attention since it does not fit with the other topics at all.\u201d<\/span><\/p>\n This response, Alex added, meant that Opus didn\u2019t just find the \u201cneedle\u201d, but correctly identified it as being placed in the \u201chaystack\u201d as a test.<\/span><\/p>\n \u201cThis level of meta-awareness was very cool to see but it also highlighted the need for us as an industry to move past artificial tests to more realistic evaluations that can accurately assess models true capabilities and limitations,\u201d Alex said.<\/span><\/p>\n So, only slightly terrifying then.<\/span><\/p>\n Featured Image: Photo by Aideal Hwa<\/a> on Unsplash<\/a><\/em><\/strong><\/p>\n<\/p><\/div>\n\n
Needle in a haystack<\/h2>\n