my neighbor started a blockchain trading company. (Im going to be cryptic about what it does). He fired all the programmers who didn’t embrace AI to help their jobs. He has less headcount and more productivity.
Those programmers he fired were probably useless anyway. In a typical team of 10 programmers, only 3 or 4 will be really good. Seen it at almost every company i worked for. This applies to other departments and fields as well, not just programming. This is how Musk fired 80%+ of twitter and it still worked, perhaps even better now. Some programmers are even net negative on the team, they submit buggy code and need lots of help from experienced devs. And middle management rarely fires them as long as the good devs get the work done.
You could also look at this the other way with how hard it has been to do better than chatGPT4. Claude 3 may have gained an edge at this point but it isn't even enough to get me to switch after a year of chatGPT4. I would have thought a year ago that I would be using something more advanced from Google than chatGPT4 by now. There is also such different experiences depending on what programming language we are talking about. I don't know what Python function I wouldn't be able to write with chatGPT4. R is almost useless to even bother with for chatGPT4. ChatGPT4 can do amazing things writing javascript and python but that is because of the training data and not the model. The STAN programming language on the other hand it can write the most basic program but is entirely useless. It seems like we have clearly over estimated the trajectory up from here when it has taken the entire internet's worth of training data to get to this point. I just can't imagine there is not the worst AI winter of all time between here and AGI. It is also shocking really the lack of any kind of useful product built on top of the APIs when it seems like almost the entirety of the computer industry is racing to figure out what that product is and to build it.
anyone have an example of LLM generated code running in production? I know it is used to generate chat, articles, ads, ... but haven't seen an actual software application. Also, if an LLM can generate working software would it just produce the text or can it test, compile, deploy, monitor, ...
I used to occasionally hire Python and R programmers from Fiverr and Upwork in the past years. Nowadays I use ChatGPT to spit out code for me rendering the programmers unnecessary. I'm a very small example but it's got to be happening on a much larger scale.
A month ago an AI agent called 'Devin' came out to lots of buzz and hype because it claimed to be the first 'AI software engineer'. You can watch videos on youtube, but it can supposedly write,compile,test,debug,deploy etc. Its an agent that has its own development sandbox, I think it uses GPT 4.0 underneath. But the company that developed it is still actively hiring 'Human Software engineers' on its website, so don't believe all the hype... This guy makes claims of 'long term planning and reasoning' breakthroughs, but sounds like BS to me if he is just using ChatGPT underneath.. This is a typical AI marketing video, they always make it look like everything works perfectly..
Would you care to post something that you would like ChatGPT to build? And then post the code? I am curious about what might come out.
sure. Here was the question I posed to GPT: dataframe RawStream has columns DateTime and Price. dataframe EarlyCloseDates has a list of dates. For dates that are in EarlyCloseDates, I want to filter out data after 1pm and here was GPT's first response: To achieve this, you can first convert the 'DateTime' column in the RawStream dataframe to a pandas datetime object. Then, you can merge the dataframes based on the 'DateTime' column and filter out the rows where the date is in the EarlyCloseDates dataframe and the time is after 1 pm. Here's how you can do it: import pandas as pd # Assuming RawStream and EarlyCloseDates are your dataframes # If not, replace them with your actual dataframe names # Convert 'DateTime' column to datetime format in RawStream dataframe RawStream['DateTime'] = pd.to_datetime(RawStream['DateTime']) # Merge dataframes based on the 'DateTime' column using a left join merged_df = pd.merge(RawStream, EarlyCloseDates, left_on=RawStream['DateTime'].dt.date, right_on=EarlyCloseDates['Date'], how='left') # Filter out rows where the date is in EarlyCloseDates and the time is after 1 pm filtered_df = merged_df[(merged_df['Date'].notna()) & (merged_df['DateTime'].dt.hour < 13)] # Drop the additional columns introduced by the merge operation filtered_df = filtered_df.drop(['Date'], axis=1) # Print the resulting dataframe print(filtered_df) This code first converts the 'DateTime' column in the RawStream dataframe to a datetime object. Then, it merges the dataframes based on the date, filters out rows where the date is in EarlyCloseDates and the time is after 1 pm, and finally drops the additional columns introduced by the merge operation. Adjust the column names and dataframes based on your actual data, and this should filter out the data as per your requirements.
With all due respect. That is a very very easy task. You just have to google how to filter data with pandas and it will point you to stack overflow where you can copy the code. In fact, I wouldn't be surprised if that is what ChatGPT is doing, scrapping the web for code samples.