- The Brainyacts
- Posts
- 108 | š¦ š§æ Claude2 Takes the Lead
108 | š¦ š§æ Claude2 Takes the Lead
Brainyacts #108
In todayās Brainyacts we:
replace OpenAI with Anthropicās Claude2
talk Googleās new Bard features
marvel at a new Google notetaking tool
cover LexisNexisās recent GenAI survey
explain what GPUs are (compared to CPUs)
learn OpenAI and Associated Press team up
look into FTC investigation of OpenAI
wonder if trolls will use AI to badger companies
ask if patents will get harder to secure
look at Chinaās new generative AI rules
cover allegations that Google stole data to train AI
share more news you can use
š to new subscribers!
To read previous posts, click here.
šøļø š° Anthropicās Claude2 is simply amazing
A new free conversational AI tool just dropped, and it is huge!
Anthropicās Claude2 was released earlier this week and it has significantly raised the bar for all other chat-based tools.
Here are some of its features/vitals:
ā¢ You can upload files (5 at 10MB each) and have it review them
ā¢ It can digest up to 75,000-word prompts, compared to ChatGPT's 3,000-4,000 word limit (OpenAI Plus gives you a bit more, but not much)
ā¢ It knows things from early 2023 versus Sept. 2021
ā¢ It appears to be fast (but as its popularity grows, we shall see)
ā¢ It is only available in the US and UK right now
You can sign up here.
You can read the underlying paper on Claude2 (worth it IMO) here.
Here is a short video from Anthropic covering its ability to ingest and read multiple PDFs.
And here is a short video I made showing the strengths of Claude2 in one use case.
š¦ š Google Bard: From Embarrassment to Excellence
Google Bard is an AI chatbot like ChatGPT, Claude, and Bing Chat.
Bard is powered by a large language model (LLM), a prediction engine that generates responses based on words likely to come next. Bard uses a lightweight and optimized version of LaMDA, Googleās LLM (different from OpenAIās GPT-4).
Bard sounds amazing, right? But it wasnāt always so. When Google first launched Bard in early 2023, it was met with criticism and ridicule. Many people found Bardās responses to be inaccurate, misleading, biased, or nonsensical. For instance, Bard gave wrong scientific names for plants, confused facts about history and geography, and made offensive jokes and stereotypes. Some users even tried to trick or troll Bard into saying absurd or harmful things.
Google acknowledged the challenges and limitations of LLMs and admitted that Bard was still an early experiment that needed more feedback and improvement. Google also added features to Bard to make it more helpful and reliable, such as giving multiple drafts of responses for users to choose from, allowing follow-up questions and alternative requests, and grounding responses in quality information from the web.
Since then, Google has been working hard to improve Bardās quality, speed, and safety. Google has also been expanding Bardās availability to more countries and languages, and inviting more users to try it out and share their feedback. Google has also been showcasing Bardās potential and benefits for productivity, creativity, and curiosity.
Today, Bard is no longer an embarrassment, but an excellence.
Yesterday, they released some amazing updates:
Bard is now available in Europe and Brazil
Bard can now talk, with text-to-speech
Truly amazing, Bard can see!
Letās dig into that text-to-speech first.
To test this, I asked it to help me translate something from English into German. It worked perfectly. Check out this short video demo.
Now letās look at the visual ability.
You can now upload a picture or graphic directly in Bard and ask it questions about it.
Here is my first test of this:
That was really great, I think. So, I tried to push it further by taking a picture of my chicken scratch notes of mine to see if it could recognize and decipher them. Now these are hard-to-read notes that I even struggle to figure out what I was writing. But look š, Bard does a fairly great job at it. Truly cool!
š š¤ More Google AI Tools
If you use Google Docs, this is for you.
Something intriguing just popped up from Google Labs. We all have struggles managing the massive amounts of information we deal with daily. Well, Google's been working on a solution called NotebookLM, and it's shaping up to be quite the game changer.
So, NotebookLM is a notetaking tool, but with a twist. It's essentially an AI-powered notebook, a part of Google's Project Tailwind, designed to help with understanding and processing loads of information. But it's not just your run-of-the-mill AI. You can feed it your notes and Google Docs, and it learns from them, essentially becoming a personalized AI assistant.
Once you plug your Google Docs into it, the AI provides a summary and even suggests potential questions to help you get a clearer picture of the material. You can also ask it to delve deeper into the material, like having it explain certain terms or summarizing specific events in a text.
What's really fascinating is that it's not just about question and answer with this tool. NotebookLM can assist with brainstorming new ideas, too. Imagine if you're working on a project and need some creative input, you can ask NotebookLM and it will generate some fresh perspectives.
And to top it off, it always backs up its responses with references from your original documents, making it easier for you to cross-verify the info.
You need to get on the waitlist from Google Labs - do that here!
š š¤ LexisNexis just released an incredibly helpful survey, āGenerative AI & the Legal Profession.ā
The survey, conducted among 1,175 UK legal professionals from May to June 2023, finds 87% of legal professionals are aware of generative AI tools and of that group, 95% agree these tools will have an impact on the practice of law (38% said it will have a significant impact,11% said it will be transformative and 46% thought it would have āsome impactā).
I highly recommend you download a copy and check it out for yourself
Here are some great insights.
ā”ļøš» What are these things called GPUs?
Under the Hood of Generative AI: The Power of GPUs
Most people are familiar with the CPU, or central processing unit, which is the "brain" inside computers and mobile devices that processes instructions and runs programs. CPUs have a few cores optimized for sequential, generalized tasks - like running a web browser, word processor, or operating system. Less known is the GPU, or graphics processing unit. Originally designed to handle complex graphics and visuals (heavily used in gaming), GPUs have proven remarkably adept at accelerating AI as well.
To understand the difference, consider a simple task like making breakfast.
š³ A CPU would handle this sequentially, first cracking several eggs, then frying some bacon, then toasting bread - finishing one step before starting the next.
š³ š„ š„Æ A GPU, with thousands of small cores, would have some cores cracking eggs, some frying bacon, and some toasting bread - all at once in a massively parallel fashion.
GPUs can contain thousands of these small processing "cores" that operate in parallel, handling many calculations simultaneously. This makes them ideal for the math-heavy matrix operations underlying neural networks, the engine of generative AI models like DALL-E and GPT-4. CPUs, with fewer, more general-purpose cores, are optimized for the sequential processing of instructions in the programs we run every day.
We're still early in unlocking the potential of GPUs for AI. Today's models use GPUs inefficiently, unable to fully utilize the thousands of cores. With optimized software and workloads, we are learning to get far more out of each GPU. For example, Anthropic, designed its Claude chatbot to better leverage GPU parallelism, achieving state-of-the-art results with 10x fewer parameters than its rivals. You can see/feel the difference!
Better utilizing GPUs will make generative AI more efficient and accessible. Models that tap into the full power of GPUs require fewer parameters and less data to train, lowering compute costs and reducing demand for scarce GPU hardware. Startups will find it far easier to develop and deploy advanced AI when a single GPU can achieve what once required a prohibitively expensive cluster. End users will benefit from AI that runs locally and privately on their phones, tablets and computers. Unlocking the vast potential of GPUs may be the key to bringing the power of artificial intelligence to everyone, everywhere.
News you can Use:
In the Memetime:
Not a meme per se, but it seems like a version of one.
Use AI to generate excuses when you make a mistake at work.
Was this newsletter useful? Help me to improve!With your feedback, I can improve the letter. Click on a link to vote: |
DISCLAIMER: None of this is legal advice. This newsletter is strictly educational and is not legal advice or a solicitation to buy or sell any assets or to make any legal decisions. Please /be careful and do your own research.8