The former Openai engineer describes what is really like work there
Three weeks ago, an engineer named Calvin French-Awen, who worked on one of the most promising new products in Openai, resigned from the company.
It has just been published Wonderful blog As it was at work there for a year, including the enemy without sleep to build the manuscript. this The new coding agent from Openai That is competing with Tools like an indicator And the anthropologist Claude icon.
French Owen said that he did not leave because of any “drama”, but because he wanted to return to the founding of the startup. He was a co -founder of the customer data starting, That Twilio bought in 2020 for $ 3.2 billion.
Some of what he revealed about the culture of Openai will not be surprised, but other notes fight some misconceptions about the company. (It is immediately accessible to comment.)
Fast growth: Openai wrote from 1000 to 3000 people in the year in which he was there.
The LLM maker is definitely reasons for such employment. It is the fastest growing consumer product ever, and its competitors are growing quickly. In March, he said that Chatgpt had more than 500 million activists Users and climbing quickly.
chaos: “Everything is broken when you expand quickly: how to communicate as a company, reporting structures, how to charge the product, how to manage and organize people, employment, etc.”, French-AWEN Books.
Like a small emerging company, there is still enabling people to work on their ideas with a small red strip. But this also means that the multiple difference repeats efforts. “I must see half a scale of libraries for things like managing the waiting menu or the agent’s episodes,” showed as examples.
The coding skill also differs from the seasoned Google engineers who write code that can deal with a billion users, to the newly backward doctorate. This, in addition to the flexible snake language, means that the central code depot, nicknamed “The Background Melivery”, is “a little dumping land”.
It often breaks things or can take excessive time to operate. But senior engineering managers realize this and work on improvements.
“The launch of the soul”: Openai does not seem to know after it is a giant company, all the way to run completely on Slack. He stated that it is very similar to Move -f Break-THINGS META in its early years on Facebook. The company is also full of appointments from Meta.
French Owen described how his older team is about eight engineers, four researchers, designers, two of the going to the market and a product manager designed and launched at Codex in just seven weeks, starting to finish, almost without any sleep.
But its launch was magic. Only by running, they got users. “I have never seen a product that only gets an immediate height of appearing on the left sidebar, but this is the Chatgpt power.”
Usage Aquarium: Chatgpt is a very high company. This led to a culture of confidentiality in an attempt to leak on the leaks for the public. At the same time, the company is watching X. If a virus post goes there, Openai will see it, and may respond to it. “A friend of mine joking,” This company is working on Twitter feelings, “wrote.
The biggest wrong idea: The French Owen implicitly that the biggest wrong idea of Openai is that he is not interested in safety as it should. Certainly, many of the people of the safety of artificial intelligence, including the former Openai employees criticized her operations.
While ISIS is a theoretical risk towards humanity, there is a greater focus on practical safety such as “hate speech, abuse, manipulation of political biases, formulation of biological weapons, self -harm, and rapid injection.” He wrote that Openai does not ignore the possible long -term effects. There are researchers looking at them, and it realizes that hundreds of millions of people use LLMS today to get everything from medical advice to treatment.
Governments are watching. Competitors (and Openai scenes watch competitors in return). “The risks are really high.”
