In this CodeLetter we go over the release of Meta’s new AI frontrunner, Llama 2, Epic Games new service that hopes to dominate the gaming retail space, and the continued optimization, and backlash, of improving AI models today.

Meta Releases Llama 2, Their GPT-4 Competitor, To Developers For 100% Free.

Meta’s new large language model, Llama 2, has quickly gained traction in the AI world and is already at the forefront alongside ChatGPT and Google’s Bard. This comes with the release of Llama 2 in late July, where Meta created the service with open source capabilities and free for commercial use. This is the first AI model to hit the market that is completely free for developer use.

As innovative and efficient as Llama 2 has shown to be, questions have arose surrounding how this AI model became so efficient so quickly. As of now, the information surrounding training data and documentation is extremely sparse, with the only indication of how it was trained was through “publicly available online sources”. Developers from other AI companies became wary of these details, with concern some of this data could have come their models themselves.

With these details Llama 2 has received a polarizing, but primarily positive, response from the tech community. While details are lacking on the backend of the service, the notion of a free, open-source AI model with developers in mind is extremely enticing for programmers and entrepreneurs. (Source)

Platform Wars - Steam Facing Competition As Epic Games Store Launches 100% Revenue Cut In Favor Of Developers

Epic games has recently announced a new incentive plan for developers called First Run. This plan puts game developers first, allowing them to keep 100 percent of their games revenue for the first 6 months of the games release. This agreement requires the game to be exclusively released to the Epic Games Store as opposed to competitors such as Steam or Google Play. However, there is some flexibility with developers being allowed to simultaneously release their games on their own personal site or store alongside Epic’s.

This level of compensation split with developers and distributors has yet to be seen in the gaming industry with Epic aiming to set a new standard for developer compensation. The First Run is another of Epic’s ways of bringing in customers on top of their current weekly free game service they promote on their store.

Steam (A competitor in the space) takes close to 30 percent of revenue with added fees throughout a games lifespan. Whether Steam will make a change to their current policies or implement a similar service is unknown at this time. (Source)

OpenAI Launches System To Easily Fine-tune GPT-3.5 With Custom Data

The creators behind ChatGPT, OpenAI, have implemented a new feature for fine tuning their GPT-3.5 service with custom data. This opens the doors for developers to implement their own data and get better feedback from the AI for their specific use-cases. Custom model scalability has been increased with the addition of this feature allowing for more dynamic iterations and responses.

OpenAI further improves the service by allowing users to customize the tone of the responses the AI returns. Improved steerability and formatting ensure that generated responses are further fine-tuned and catered to the users specifications. The service encourages developers to utilize this these new features alongside function calling and prompt engineering for the best, most optimized results.

The AI company touts the notions that GPT-3.5, with these new improved features, now matches GPT-4 on various level capabilities and tasks (Source).

Stack Overflow's Resistance Against OpenAI's Utilization of Forum Posts in AI Training

Stack Overflow, one of the leading sites for developer Q&A, is joining the push to be compensated for their data being used on AI platforms. As of now, web-scraping has been the main practice of large scale AI companies such as Google and OpenAI in order to acquire training data for their services. This has led to push back from the companies this data is being pulled from.

Reddit has already announced they will start charging companies for using web-scraping on their site. This is in an effort to receive fair compensation and usage of their data.

“Community platforms that fuel LLMs absolutely should be compensated for their contributions so that companies like us can reinvest back into our communities to continue to make them thrive,” Stack Overflow’s CEO Prashanth Chandrasekar says. “We're very supportive of Reddit’s approach.”

Stack Overflow and Reddit have yet to release pricing information or specifics on their approach to receive compensation for the usage of their data. Questions about potential user compensation for community posts and updates have arisen as well. It is to be seen how these companies will handle this push going forward but updates will likely come about in the following months (Source).

Thanks for reading this weeks CodeLetter, same time next week ☮️ . If you enjoy reading CodeLetter, refer a friend by sending them over to thecodeletter.com!

- The Team at CodeLetter