From: mk_thisisit
Open AI is a company known for developing advanced artificial intelligence, including ChatGPT [00:00:58]. Representatives from Open AI visited Warsaw as part of an initiative implemented in cooperation with NCBR ideas [00:00:50]. The company aims to ensure that the benefits of its technology flow to all people [00:17:12], with a non-profit part playing a leading role to uphold Open AI’s mission [00:17:15].
Key Polish Individuals and Their Journeys
Szymon’s Journey to Open AI
Szymon, one of the creators of ChatGPT, shared his path to joining Open AI [00:01:04]. Initially, after studying at Cambridge, he was skeptical about artificial intelligence, seeing “not much hope there” [00:01:17]. He then shifted his focus to normal computer science and distributed systems at MIT [00:01:36].
While at MIT, Szymon met a friend interested in neural networks, which, despite initial skepticism, seemed technically interesting to him [00:01:38]. He began exploring how to make neural networks operate faster across multiple computers [00:01:48]. A pivotal moment was the publication of the AlphaGo paper, which demonstrated significant progress in AI [00:01:57]. This work showed that AI could achieve more than just ordinary search, like Deep Blue in chess, by representing intuition [00:02:23]. Szymon observed that these networks could internalize implied data, allowing them to draw conclusions beyond superficial information, as seen in Go where the model could play games it hadn’t explicitly seen [00:03:38]. He realized “some real intelligence” was present, indicating potential for significant progress [00:02:37].
Leadership of Jakub Pochocki
Jakub Pochocki, a Pole, led the GPT-4 project [00:05:05]. He also worked extensively on the GPT 3.5 project, which formed the basis for the original version of ChatGPT [00:05:10].
Core Contributions to GPT Projects
Team Involvement and Optimization
The development of ChatGPT was a massive project involving many contributors [00:04:51]. Polish programmers, including Szymon and Jakub Pochocki, played a significant role in optimizing the basic model that powered ChatGPT [00:05:16]. Szymon describes his role as an engineer and scientist, focusing on optimization and distributed systems [00:18:39]. In the GPT-4 project, he worked closely with Jakub Pochocki, addressing bottlenecks, adding metrics software, and solving scientific questions that hindered progress [00:18:48].
The “Voice of Polish Programmers”
Due to these contributions, it can be stated that the “voice of Polish programmers, scientists, sounds quite strong” within Open AI [00:05:29]. This highlights the substantial scientific and programming work carried out by Polish individuals.
Factors Behind Open AI’s Success
Szymon attributes Open AI’s success, particularly with ChatGPT, to several factors:
- Technological Development: A strong focus on the development of the technology itself was key [00:06:40].
- Problem Solving: The ability to solve technical and organizational problems was crucial, enabling universal access to these models for practically anyone with internet [00:07:16].
- Universal Access: The universality of access was important, as the models require extensive calculations and complex engineering solutions to be widely available [00:07:39].
- Form of Chat: The form of ChatGPT, creating the illusion of conversation with a live person, was also important [00:07:43].
Future Directions and Ethical Considerations
Open AI continues to develop its products and core scientific research to produce smarter models [00:08:42].
Developing Smarter Models
Future efforts will involve further scaling of models, as this approach has proven effective [00:09:33]. Research also focuses on addressing specific limitations, such as the probability of error in reasoning, which currently prevents models from acting as “automatic scientists” [00:09:41]. The goal is to create mechanisms that can confirm scientific claims or find proofs for mathematical claims [00:10:27]. Smarter models are also better at following instructions and adhering to imposed standards [00:14:04].
The Ethical Aspect and Public Involvement
Open AI is actively working on the ethical aspects of AI development, recognizing the need to define boundaries for artificial intelligence [00:11:32]. The company believes that decisions on AI ethics should not be made by a single company but democratically, with technology being used to collect public opinions [00:12:10]. This involves engaging as many people as possible in co-creating an ethical code [00:12:29]. The rapid growth of ChatGPT’s popularity has made solving these ethical problems urgent [00:12:37].
A technical challenge is ensuring the model respects the defined ethical standards [00:13:38]. Open AI is working on this, with Szymon personally finding that smarter models are inherently better at following such instructions [00:13:47].
Improving Polish Language Effectiveness
Open AI is actively considering improving the effectiveness of ChatGPT in Polish [00:14:41]. The GPT-4 model shows significant progress across many languages, including less common ones like Swahili, where GPT-4 performs better than GPT-3.5 did in English [00:14:59]. Polish performance is currently positioned between English and Swahili [00:15:16]. Progress is partly due to creating smarter models that can transfer knowledge from English to Polish, but more high-quality Polish data would further help reduce the language difference [00:15:26].
Open AI’s Relationship with Microsoft
Microsoft is described as a “great partner” and an investor, providing computing power necessary for developing smarter models [00:16:06]. Microsoft also assists in sharing Open AI’s technologies, such as in GitHub Copilot [00:16:22]. Crucially, Microsoft does not have control over Open AI, and does not hold a single seat on its board [00:16:31]. This structure is intentional to prevent a publicly listed corporation, obligated to investor interests, from making decisions about technology that could profoundly affect everyone’s lives [00:16:53].
Future of AI: Superintelligence
Szymon suggests it is “wise to think about a scenario” where strong artificial intelligence, smarter than some of the smartest people, could appear before 2030 [00:18:12]. Even if the probability is small, it is something to prepare for [00:18:25]. The skills of current models seem more general, prompting questions about the move towards general artificial intelligence or superintelligence [00:17:39].