AI tests - post in progress

This post documents tests which are used to assess AI LLM performance. As I find more tests and understand them I will post them here.


AGIeval is a test to see whether AIs such as yourself are able to pass human tests. (AGI evaluation). AGIEval focuses on advanced reasoning abilities beyond natural language through inductive, deductive, and spatial reasoning questions. Models must solve challenges like logical puzzles using language. It appears in a paper on ArXiv as follows: 

[Submitted on 13 Apr 2023 (v1), last revised 18 Sep 2023 (this version, v2)]

AGIEval: A Human-Centric Benchmark for Evaluating Foundation Models

by Wanjun Zhong, Ruixiang Cui, Yiduo Guo, Yaobo Liang, Shuai Lu, Yanlin Wang, Amin Saied, Weizhu Chen, Nan Duan. 

Traditional benchmarks, which rely on artificial datasets, may not accurately represent human-level capabilities. In this paper, we introduce AGIEval, a novel benchmark specifically designed to assess foundation model in the context of human-centric standardized exams, such as college entrance exams, law school admission tests, math competitions, and lawyer qualification tests...


MMLU stands for  Measuring Massive Multitask Language Understanding and it is a speedtest/benchmark for LLMs. It appears on ArXiv as follows: 

[Submitted on 7 Sep 2020 (v1), last revised 12 Jan 2021 (this version, v3)]

Measuring Massive Multitask Language Understanding

Dan Hendrycks, Collin Burns, Steven Basart, Andy Zou, Mantas Mazeika, Dawn Song, Jacob Steinhardt

We propose a new test to measure a text model's multitask accuracy. The test covers 57 tasks including elementary mathematics, US history, computer science, law, and more. To attain high accuracy on this test, models must possess extensive world knowledge and problem solving ability...

Popular posts from this blog

Testing ChatGPT on coding - a script to make CSV to ICS

 So I gave ChatGPT a task of writing a CSV to ICS convertor, to take a spreadsheet of events and turn it into individual calendar file invitations. The idea is you can make a calendar with lots of event items in a spreadsheet, and then output ICS that you can just double-click to add to your calendar automatically. So here's the spreadsheet: "Event Name","Event Start Date","Event End Date","Event Start Time","Event End Time","event description or notes","URL applicable" "Meeting with John","2024-06-10","2024-06-10","09:00","10:00","Discuss project updates.","http://example.com" And here's the script ChatGPT made. It seems to get confused with things like Daylight Savings and/or Timezone but you can fix that manually. It took some debugging and I had to copy/paste my Mac's manpage for DATE into ChatGPT, but it got it eventually. The only
Read more

Ollama3 works - running a local LLM

Do this: curl -fsSL https://ollama.com/install.sh | sh ollama create llama3  It will then download the llama3 model, about 4.7 gb to run it: ollama run llama3  if you want it to read and summarise a textfile called textfile.txt: ollama run llama3 "Read the text file $(cat textfile.txt) and summarise the findings in one paragraph of no more than 300 words"  And here's a gui for it: https://github.com/amithkoujalgi/ollama-pdf-bot The system runs as a service (daemon) on Linux. CPU usage is high but not bad and response time much better than localGPT. It stores the model files (large LLM blobs) in  /usr/share/ollama/.ollama/models The models can be dowloaded from huggingface.io or https://ollama.com/library
Read more

What is a "token" , what is "temperature", and what is "n-shot"?

 TL;DR: token is a meaning unit; temperature is randomness. Token: In the context of AI and natural language processing (NLP), a "token" refers to the smallest unit of text or language that a model can understand. Tokens are usually words or subwords, but they can also be characters or even smaller units, depending on the tokenization method used. Word Tokens: In most cases, a token corresponds to a word in a sentence. For example, the sentence "ChatGPT is helpful" would be tokenized into three word tokens: ["ChatGPT", "is", "helpful"]. Subword Tokens: Some models use subword tokenization methods like Byte-Pair Encoding (BPE) or WordPiece to split words into smaller units. For example, "unhappiness" might be tokenized into ["un", "happiness"]. Character Tokens: In some cases, especially for languages with complex characters or when dealing with very short text, tokens can be individual characters. For ex
Read more