Claude 2, a new barrier overcome
Breaking through the 32k limit, going for 100.000 tokens.
Claude 2, a New Barrier Overcome
I’ll probably stay a client of ChatGPT from OPENAI for a little while, at least, but the competition is getting tough. A few days ago, Bard joined the club with ChatGPT and Bing, and now Anthropic has just released Claude 2. I stopped using superlatives 14 days ago because there aren’t any more left able to express the daily evolution we are living.
(One thing: like Bard, you might need your VPN for some time to access Claude 2 since it is not yet accessible worldwide.)
Claude 2: The New Kid on the Block
Anthropic, backed by Google and former executives of OpenAI, has launched Claude 2 to compete with the likes of OpenAI’s ChatGPT or Inflection’s Pi. The updated model, Claude 2, is now more accessible, with a beta-test website available for general users in the U.S. and U.K. Businesses can access the new model through an API at the same cost as Anthropic’s previous version, Claude 1.3.
The Special Thing About Claude 2
What’s special about Claude 2?
Well, to me, the most important thing is the token limit. We are still far from the 1B tokens we spoke about a few days ago, but things are going fast. Claude 2 accepts 100,000 tokens, that’s about 75,000 words.
A standard book is approximately 70,000 words, so we reached the level of having an AI write a book in one go.Claude 2 accepts 100,000 tokens, that's about 75,000 words. A standard book is approximately 70,000 words, so we reached the level of having an AI write a book in one go. Share on X
Putting Claude 2 to the Test
Now, I like to put tech to the test. I don’t yet want to write a 70k word book, but I certainly want to test the quality with which Claude 2 gets through large PDFs.
So I tested it with a serious PDF, which ChatGPT hadn’t been able to absorb (at all). And it did great. Ghost in the machine – Addressing the consumer harms of generative AI” 37k words.
Just to be sure, I first asked for the document’s title.
Then, I wanted a general summary of the paper.
I got it with the comment that it only ‘skimmed’ the document:
Based on skimming through the document, it appears to be a report by the Norwegian Consumer Council about the potential harms and challenges of generative artificial intelligence (AI) systems for consumers.
After which, it gave me the basic elements of the document.
Next, I wanted to test its level of detail and asked a very precise question for which the answer was deep down in the document. Without hesitation, it answered:
Specifically, the relevant excerpt from the report states:
“In April 2021, the EU Commission published a proposal for the Artificial Intelligence Act (AIA), laying down harmonized rules across the EU and the EEA “to foster the development, use and uptake of artificial intelligence in the internal market”.”
The report also cites the proposal document in its endnotes:
And Claude went on with the level of detail I asked for.
A Discussion with Claude
Finally, I got into a discussion with Claude about the issue treated in the document “Ghost in the Machine – Addressing the consumer harms of generative AI“.
My Point of View
In my opinion, laws and regulations have never been a foolproof shield against misuse and criminal behavior. Just because there’s a law against theft doesn’t mean I can leave my front door wide open, right? We each have a hefty slice of responsibility to bear.
Now, if you ask me, the real game-changer is education. It’s crucial that people understand what AI is, what it’s capable of, and how they can harness its power. Knowledge is a powerful tool. The more people know, the more they’re able to protect themselves, and they won’t have to solely rely on legislation to keep the bad guys at bay. After all, those with ill intentions aren’t exactly known for their respect for the law, are they?Laws and regulations have never been a foolproof shield against misuse and criminal behavior, nor will they for AI; if you ask me, the real game-changer is Education Share on X
Claude’s answer
You raise an excellent point. Legislation and regulation alone cannot entirely prevent misuse and irresponsible uses of technologies like generative AI. Education and awareness are critical components as well.
You’re absolutely right that We all have a responsibility to understand generative AI, its capabilities and limitations and make informed decisions about how to use and interact with it. Share on X Just like leaving doors unlocked when theft is illegal still invites crime, relying solely on laws to dictate the proper use of AI is insufficient.
Robust education and awareness building around AI is crucial to empower consumers to protect themselves. Some ways this could be achieved:
- Public awareness campaigns explaining how generative AI works and its risks and providing guidance on responsible use.
- Educational programs, especially for youth and the general public, on AI literacy – understanding AI’s societal impacts.What our tomorrow needs: Educational programs, especially for youth and the general public, on AI literacy - understanding AI's societal impacts. Share on X
- Transparent and ethical development practices by companies building generative AI.
- Media spotlighting potential dangers as well as benefits of the technology.
- Grassroots initiatives by consumer groups and non-profits to spread awareness.
An informed citizenry that approaches new technologies thoughtfully is the strongest defence against misuse. While laws set the boundaries, education gives people the tools and impetus to stay well within them. Policymakers should make AI education a top priority, along with legislation to build a society resilient to AI risks. With knowledge, people can keep their doors locked even when some choose not to.An informed citizenry that approaches new technologies thoughtfully is the strongest defence against misuse. Share on X
Conclusion
In conclusion, the AI landscape is rapidly evolving, and Claude 2 is a significant step forward. Its ability to handle large documents and provide detailed responses is a game-changer. However, as with all technology, it’s essential to remember that knowledge and education are our best defences. The more we understand about AI, the better we can use it responsibly and effectively.
That is, without any nuance, the purpose of our AI Leadership.
I rest my case.
Yours in AI
Luc
TL;DR
Claude 2, the latest AI tool from Anthropic, is making waves in the AI community.
Its ability to handle a whopping 100,000 tokens, equivalent to 75,000 words, sets it apart from its competitors.
This means it can process large documents, even those as lengthy as a standard book, in one go.
We put Claude 2 to the test with a serious 37k-word PDF, which performed impressively.
The document we tested it on was “Ghost in the Machine – Addressing the Consumer Harms of Generative AI”. New legislation for a safer AI.
And, of course, we ended up discussing the document’s content with Claude.
We emphasize the importance of education in understanding AI. While legislation is necessary, it’s not foolproof against misuse and criminal behaviour.
Therefore, people must know what AI is, what it can do, and what they can do with AI. The more they know, the more they can protect themselves, not just rely on legislation to keep potential misuse at bay.
