Claude Data Analysis Just Got a Major Upgrade

[@SkillLeapAI]

Join the fastest growing AI education platform and instantly access 20+ top courses in AI: bit.ly/skillleap

[@beneditofredericocongolojr1567]

Hi, Love your videos. I have a question: what's the context window for Claude's Professional Plan? How many documents can you attach and its respective size limit. If you could express the context window in page numbers I'd appreciate it. And I have an idea for a new video - "Claude PRO vs Perplexity Spaces - Pro's and Con's"

[@analysisjitsu]

To work around context limit you could (1) upload a subset of the data, (2) ask it to write python code to analyze the data, from a csv, (3) ask it to write code (python or react) to visualize the data, and then (4) run the analysis/visualization on the whole dataset from your local computer.

[@user-xr4hx4hx9m]

Yeah, that's what I was thinking too.

[@ricardocnn]

Claude is by far the best model

[@YTBenZ]

Thank you! Interesting that your 2nd highest viewership city is Bengaluru.

[@SkillLeapAI]

yea never knew that

[@canastasio29]

Thanks as always, Saj!

[@roxnroll8050]

Wow! It's so close on predicting Bitcoin at 97K! Great video

[@jeremyfmoses]

The needed context for most .csv files will likely only be the first line, and then the rest of the file should only be accessed by the generated Javascript code. I suspect the context window limitation can be bypassed by one of three things: 1) giving Claude only the first few lines of the .csv and then copying the resulting Javascript for use outside Claude, 2) some sort of clever prompt engineering, or most likely 3) Anthropic will address this limitation specifically in the next version of Analysis, probably by ignoring all but the first few lines of the .csv.

[@paulmiller591]

Great overview thanks!

[@iDannyism]

did you verify any of the outputs for dataset accuracy?

[@AI_Revolution13]

I love your videos Sal, but this is a bit off topic. I like the look of red circle on your pointer, is there a way I can enable that on a Mac? Thanks.

[@bondnikunj]

Can ai answer this? Maybe send a screenshot and ask the tool name

[@SkillLeapAI]

Hey thanks. Yea there is an app called pointer pro that I installed. I think it’s 3 bucks on the App Store

[@opita]

BTC price projection was spot on. Hopefully it didnt try to pull from current knowledgebase.

[@duckrabbithead]

Does this type of update essentially mark the beginning of the end for jobs like business analyst/data analyst? If something like this could connect to larger files/databases, and either connect to data visualization tools like Tableau or use its own data viz tool, that would replace a huge part of those jobs. I'm no AI expert, just a Business Intelligence Analyst following the trend, thanks!

[@timooothy1234]

You supposed to use it to elevate your workflow, the only thing replacing those kind of jobs are people who embraced it and are using it not AI itself

[@duckrabbithead]

@@timooothy1234 I get that its a productivity enhancement tool right now. But if the tool (AI in general) continually improves like it has in past few years, how soon before a manager/executive simply has to ask it a series of questions for data analysis/presentation without needing a human data analyst? or at least not a team of analysts? I wonder how easy/hard that capability is for things like Claude or o1. Are companies like Anthropic actively working on that ability? @SkillLeapAi thanks,

[@duckrabbithead]

@SkillLeapAI

[@carlkim2577]

How did you download all those metrics from your youtube channel in one go? I would like to do the same and analyze.

[@SkillLeapAI]

It was different documents. If you go to analytics, you can click on advanced analytics. Each tab will have a download link on top of the page

[@carlkim2577]

@@SkillLeapAI thanks!

[@micbab-vg2mu]

it looks great:)

[@УмидМусаев-й5ч]

Insane! Is there any news about availability of this feature through api? Could not find anything related, and their api didnt update since 22 october

[@SkillLeapAI]

I didn’t see any API mention with this release

[@jimmycylon9437]

How is this compared to OpenAI assistant code interpreter?

[@SkillLeapAI]

pretty similar in my experience on the analysis side

[@SoloTravelGuyNow]

How can we prevent this from happening? I pay for this and very quickly get this message..."Message limit reached for Claude 3.5 Sonnet"

[@analysisjitsu]

coo. nice video

[@totomomo18]

can you use claude to upload an excel file xls/xlsx and have clause analize the file it and modify it add rows columes and even making new sheet in the file and in the end supply me the modified excel file to download ?

[@orthodox_gentleman]

You would need to convert it to CSV…OpenAI allows Excel files.

[@orthodox_gentleman]

Or you would have to use API and agentic framework.

[@CaribouDataScience]

Do you need a paid account?

[@SkillLeapAI]

They said all Claude accounts during this beta preview

[@Avegeania]

I've been using gpt and claude side by side for my algo trading personal project. But i don't always get what i intended, but have to tweak something manually to make it work. They get smarter everyday? Yes. Errorfree? Hell No.

[@maikvanrossum]

Cool… but that context window doesn’t make sense at all. It’s almost like paying for a trial version of an AI tool 🤷🏼‍♂️

[@SkillLeapAI]

I feel like lot of Ai tools feel like they are in beta. The context window is a big problem right now

[@maikvanrossum]

@@SkillLeapAI yes, you’re right… and this of course is still a beta feature. I think I’m becoming a bit sceptic and a bit disappointed, because of the quick release of multiple ‘limited’ features in such a short period of time. Most of the time the limitation is the ‘context window’. This could be due the power needed to run these kind of things, especially when most of it is for testing and/or trial purposes. At the end we will get multiple premium tiers, containing different feature sets. In that case I think Google is operating a bit more ‘traditional’ and therefore I can accept it as fair; all of their beta testing is done through the free tier. And yes, we can argue about the performance of their current model, so that’s why I’m not willing to pay for it, yet. But for instance NotebookLM was entirely tested in beta (like Gmail, Docs, etc. by users on the free tier), and now it’s released under the payed Workspace tiers, SLA, and additional Gemini Advanced license.

[@I_Mackenzie]

Interesting. I'll give this a go. Claude is still not very accurate in counting characters. Get's it wrong almost every time.

[@dataorbit27]

But it is great at writing a script that can count characters. You seem to be using a drill like a hammer.

[@timooothy1234]

It's training to build, calculate at some degree and write counting characters is something insignificant for it to learn (in my view)

[@I_Mackenzie]

@@timooothy1234 Unless what you get it to do requires a certain number of characters, then it doubles your work. Not sure I can be filled with confidence if it gets simple things wrong.

[@Grace-pm7ds]

We need a hundred percent accuracy with everything , not just financial data. That's the problem with the AI's.They're not all a hundred percent accurate and when learning under false pretenses it's only gonna spread false information. Some of these ai's, you can ask them the identical same question a hundred times and get a different answer ten of the times. While that's 90% and still an A. That's not accurate. And that's not acceptable.

[@orthodox_gentleman]

I think you don’t know fully what you are talking about.

[@Grace-pm7ds]

@@orthodox_gentleman Try it.

[@Siewe_Francois]

AI is not expected to do everything.

[@Herobrine633]

@Grace-pm7ds then you have to make your tool by your own 😂

[@Grace-pm7ds]

@@Siewe_Francois What it does do should be a hundred percent accurate

[@SoloJetMan]

I'm 12yrs old watching KZbin at midnight

[@ItsYaBoY119]

Why don’t you just add it to a project??? You say all this about larger files but don’t even try the projects??!

[@SkillLeapAI]

I don’t think this new option was inside of projects yet

[@mariusnicolaefit3583]

don’t you think chat gpt will come verry fast with something like this from behind?

[@sirabe9835]

buy me one month subscription

[@Hae3ro]

Still not good enough

[@SW-fh7he]

😅