Newsletter
Open AI, one of the leading companies in generative AI today, has been dealing with a headwind of drama. While that saga is ongoing, the company Anthropic, has been making some improvements to its ChatGPT competitor. The company unveiled Claude 2.1, the next generation over Claude 2.0. This seemingly iterative update brings some massive improvements to the AI chatbot, according to a new report. This could be the company’s chance to leapfrog OpenAI while that company is stumbling back onto its feet.
If you’re curious about what’s going on at OpenAI, you can read about it and our editorial, OpenAI Situation: What’s Going On. It shows a timeline of the drama currently going on with the company.
Anthropic just unveiled Claude 2.1, and it brings some massive improvements
One major subject in AI chatbots is the number of tokens that can be processed. A token is a snippet of a word that the chatbot can analyze and understand. Every time you type into a chatbot text box, you use up a few of the tokens that the chatbot can understand.
Chatbots, just like human beings, can only process is certain amount of information at a time. According to the company, Claude 2.1 can now understand up to 200,000 tokens. That equates to about 150,000 words. So, if you want to do so, you can paste an entire novel (well, a short novel) into Claude, and you can ask a question based on every bit of information present in that novel.
This is extremely handy if you need it to analyze long and complicated documents. Most likely, your queries won’t typically contain 150,000 words. However, the important thing is having the option to.
Claude 2.1 doesn’t only improve on what information you put in, but it also improves what information comes out. Hallucinations are a major issue for AI chatbots. When a chatbot hallucinates, it gives you a wrong answer by creating the information seemingly out of nowhere.
Well, Claude 2.1, according to the company, cuts the number of hallucinations in half compared to Claude 2.0. As with every chatbot, it’s always recommended that you double-check the information generated. Claude 2.1 is twice as likely to tell you that it does not know an answer rather than just hallucinate one. Along with that, the chatbot is able to deliver 30% more accurate answers.
Anthropic does warn users that longer and more complicated queries will require more processing time from the chatbot. With the increase in tokens, people are likely to use much longer queries. It may take the chatbot several minutes to process a single response. So, if you upload an entire case document and ask Claude to analyze it, don’t refresh the page if it’s still processing after a few minutes.
There are also some new developer tools coming to Claude
If you use this chatbot for more robust reasons, the company has some useful tools that will help you with your developer needs. Starting off, there is a new workbench console available to devs. According to the company, it will let them tweak their prompts in a “playground style experience and access new model settings to optimize behavior.”
Another feature is called Tool Use, and it lets Claude “integrate with users’ existing processes, products, and APIs.” It will be able to use services and tools from other sources to create an all-around useful experience. These tools are in beta testing, so, you should expect some bugs and instability with these features. However, that’s just par for the course. In any case, it’s always recommended that you give the company the proper feedback to help improve this.