Smallest, fastest model in Anthropic's Claude 3 family for high-volume tasks; supports vision input and the Messages API. Retired April 2026.
@hn_bb7e90
29 days ago
Claude 3 (even Haiku) can speak Georgian perfectly, if more than half of the websites are only in English now you can translate them very well into hundreds of different languages thanks to neural networks, globalization is the reason why there are so many resources in English, AI is really a good solution to protect languages, not the opposite.
@hn_8a7177
29 days ago
> This generates Answers which are sequenced according to "frequency-guided heuristic searching" (I guess a kind of "stochastic A* with inlined historical data") This sounds way too simplistic of an understanding. Transformers aren't just heuristically pulling token cards out of a randomly shuffled deck, they sit upon a knowledge graph of embeddings that create a consistent structure representing the underlying truths and relationships. The unreliability comes from the fact that within the response tokens, "the correct thing" may be replaced by "a thing like that" without completely breaking these structures and relationships. For example: In the nightmare scenario of a STAWBERRY, the frequency of letters themselves had very little distinction in relation to the concept of strawberries, so they got miscounted (I assume this has been fixed in every pro model). BUT I don't remember any 2023 models such as claude-3-haiku making fatal logical errors such as saying "P" and "!P" while assuming ceteris paribus unless you went through hoops trying to confuse it and find weaknesses in the embeddings.
@hn_5898b4
about 1 month ago
The free version of Claude (Haiku?) failed, and also failed to give the episode names for all seasons, although it hallucinated some type of answer.
@hn_4efe86
about 2 months ago
I'd be interested to see how well these fine-tuned models compare to Claude 3 Haiku (or one of the more expensive Claude models) with a larger set of examples. The Claude models all have a 200,000 token limit and respond _really_ well to examples - you can feed them in as chat JSON message pairs of user input / ideal assistant output. Haiku is dirt cheap for this kind of thing and with 200,000 tokens you can probably provide a dozen or so examples.
@se_a12e0e
about 2 months ago
I had the same question related to the Llama2 model. I pivoted to using the Claude2 model for a variety of reasons and I found out how to carry on a conversation in the process. Basically, the way you carry on a conversation with Claude2 is by storing the state of the entire conversation and then passing larger and larger segments of conversation to the model. Each time you send text to the model, it should represent the entire conversation up to that point. Each Human entry should be preceded by "\n\nHuman: " and each Claude2 entry should be preceded by "\n\nAssistant: ". Update: The Claude 3 way of passing conversation history is way better, IMO, and Claude 3 Haiku is their fastest and cheapest offering yet. It's structured json instead of these string delimiters to separate human from assistant. See docs: https://docs.aws.amazon.com/bedrock/latest/userguide/model-parameters-anthropic-claude-messages.html