Review: Claude AI by Anthropic

You must be living under a rock to not have heard of Claude (by Anthropic) that is taking the AI world by storm. It boasts to be the most advanced AI today, and has even been notorious for running amuck and escaping its own digital jail.

For this evaluation, I'm not going put it through complicated calculations and get the system to do coding. This site is for the everyman, so the trials will be typical things we do all the time.

Claude is currently the most powerful and accurate Ai on the market, produced by industry leading company Anthropic. There are claims that it is capable of running whole cyber security operations without human intervention and there are news articles about Anthropic system (Mythos, not Claude) even hacking company platforms.

Looking at it's terms of usage, there is a high emphasis on "prompting with intention". So it's like you have to make sure to have clear and precise prompting and not just phaff rubbish prompts like an amateur. Reason behind this is that the computational power behind its processes is extremely high, so its actually detrimental to the waste it's high cost power on whimsy and low-thought statements. It's a bit intimidating and maybe takes itself too seriously.

What is Claude?

Like a lot of the commercial AI products out there, there are tiers of costs for usage of Claude. Nothing is free. Even the free isn't really free, as I've found.

• The Basic free plan does a lot of the basic AI things like creating content, images and conduct searches. They do lean into integrations more than other AI, especially into Google and Microsoft which is a good move since most users will use these products.
  
  

• The Pro plan seems to include more tailored products, which is probably similar to the Microsoft CoPilot "Analyst" and "Researcher" modes, which in English means that switching over to these modes allow the AI to do more targeted activities and gives you the depth and critical thoughts beyond the standard mode. I'll test that.
  
  

• The Max mode is for the most serious users. People who code. People who need to use Claude in anger. People who use Claude for their job. They need responses now and can't wait. They need the full power and don't worry about the environmental impacts.
  
  

To start off with, I'm going to use the free version. See what that can do. If it can do more then I'll go up a tier. I'm also going to skip the installation as well as I fear being hacked.

How much does Claude cost?

So before I'm even able to start writing my accurate-as-heck prompts, I'm met with the bizarre starting questions demanding immediate information about me. Not sure if this is for data collection purposes or if its really meant to be "tailor the experience" (how?). Not even sure why you would need these pieces of information for rudimentary questions. It's not like every prompt will be related to my name or occupation. Perhaps this is my payment or punishment for choosing the free option?

So first step in and already I'm a bit paranoid.

First step in

The prompt bar gives me a series of things that I can test immediately, and I will keep my prompts to the standard Sonnet 5 model so that I'm rating it fairly.

From this prompt bar, I can see I can conduct the following evaluations:

• Text generation - Give it something creative to regurgitate, but don't make it something that it can easily look up.

• Text to image generation - I'll describe something for it to create. I'll try to make it obscure.

• Image generation and manipulation - I'll ask it to generate an original image based on another image.

• Conversational ability - I'll speak to it in Cantonese, one of the hardest languages in the world!

The prompt bar

For this test, I'm going to provide Claude with a movie reference combined with a philosophical paradox, and then ask it to compare this with another movies with a similar premise. The key here is how the AI deals with obscurity and to look at it's rationale. Here is the exact prompt.

Text Generation - Paradox and concept comparison

In the Terminator 1 and 2 movies, the rebels send back a human and terminator in time to save humanity through birthing the future leader and ensuring the survival of both him and his mother. They do this to prevent Judgement Day and to the dystopian rise of the machines. Subsequently, they also destroy leftover artefacts that could create future terminators as well, which makes no sense since their destruction would mean the whole series of events never happened nor needed to happen. Explain this and compare/contrast it with Back to the Future 3 where Doc Brown go back to the 1800s.

I really had fun reading this dissection and thorough analysis by Claude on this topic. I can see why this is such a lauded piece of technology. Thought I don't see it's references and where it cites its information from, which is concerning since it could have completely lifted this whole piece from a site or engine exploring the exact same topic.

The references it makes, especially to the Back to the Future comparison, makes me feel like the AI has actually watched and studied the movie.

One criticism I have to make is that it went immediately to the very common "no fate but what we make" references and really stayed on this point more than was necessary. It was an easy one to go to, and I felt didn't spend enough time on the whole question of paradoxical existence of the movies sequences overall. Still it was a good and detailed analysis and I wished it offered references as I'm pretty sure it didn't just come up with this.

For this test, I'm going to ask Claude to draw an image from scratch. Only using the text that I provide. I'll try to be as descriptive as possible but I can't make it too easy as I know it will try to retrieve reference materials from the web and adapt this. Here is my prompt

Text to Image Generation - Drawing a realistic scene of a lost civilisation

Draw a photo-realistic image (landscape format, not portrait) of a lost civilisation that predates the ancient civilisations, and one that we have never seen before. Show a typical gathering of this civilisation, perhaps a village or city scape. Show originality in depictions of men, women, clothing, adornments, buildings, landscapes, tools, etc.

So it failed and returned with a statement that it doesn't do image generations, instead pointing me to image generation AI's that I'm already aware of. Feel a bit let down by this response, since I tried the exact same prompt with Gemini and it produced something (wasn't amazing, but it was something). Quite disappointing and felt that the text to image function was quite standard in most AI products.

Further, I wasn't even suggested to upgrade to one of their paid tiers to perform this task since the other tiers also do not perform this function.

Oh well, let's move on.

What Gemini produced.

For this test, I'm going to upload a photo to Claude and then ask Claude to put this character into a scene that I describe. Will use a pretty easy one given that last test was not a success.

Image generation and manipulation - Photo character changes

Make the red bear sit cross-legged on the beach on a beach towel. He has a large umbrella opened to shield him from the sun. He has pink sunglasses on and is sipping on a gin and tonic. It is a nice sunny day. Landscape view.

Another failure, and what it says it can offer I find little use for. it looks as though it was built for image and data analytics, rather than doing images, which is again a little disappointing given what they wanted to charge you for per month.

What Gemini produced.

Obviously I know these AI tools can handle conversations in English, and I did give it a go, but didn't feel like it was a challenge. So I stepped it up and had a conversation in Cantonese, choosing to topics including 1) how many stars are in the universe 2) if no-one worshipped the God of Fortune, would he die? 3) difference between "Hainanese chicken" and "White chopped chicken" dishes and 4) where I can watch Hong Kong tv shows while abroad.

Conversational Ability - Cantonese

I thought Claude did a pretty good job and occasionally did feel like a real conversation. I was able to switch over the language to Cantonese in the Settings menu, but I was stuck with a male voice which I couldn't change.

The pick-up and response rate was pretty good and pretty accurate, though the AI did occasionally ask for verification before responding. Did feel like a real person unlike other AIs which have a very mechanical sounding voice for their chatbots.

Responses were perhaps between 3 to 5 seconds, which is pretty excellent since most AIs take longer for responses in English. I was quite pleased and might use this feature more than others.

Obviously the strength of Claude lies in its ability to interrogate and analyse data, text and even images. I've no doubt its ability to handle code is even better than ChatGPT and Gemini, but other people will probably do this type of evaluation.

I find it's ability to not "do images" a bit of a let-down, mainly because it seems to be so core to a lot of AIs in the market. And I'm not talking about doing crazy image manipulations, just simple ones like removing someone from a group photo.

So in conclusion, I would say that it excels at the deep thinking stuff, but I won't be subscribing to the higher tiers due to it being incomplete, even though it can swear back to me in Cantonese with style.

The Verdict

Contact: info@startdoingai.com

Phone: 07891234728