OpenAI’s Diwali Week
OpenAI sure knows how to host a Diwali party, featuring not only the Coca-Cola marketing ad with DALLE.3 but also setting off a series of bombs and firecrackers in the AI landscape.
Last week, during OpenAI's first ever developer conference, Sam Altman launched GPT-4 Turbo, an upgraded version of GPT-4, that boasts a 128k context window, allowing it to process an equivalent of over 300 pages of text in a single prompt, with knowledge extending up to April 2023.
He also made a slew of announcements that covered every part of the technology, such as open source models and developer tools, where the generative OpenAI was lacking against the competition.
Killing many birds with one stone
First came OpenAI (back in 2015) and then others (Anthropic, Cohere, and Langchain, among others) followed. The uniqueness of other platforms are based on the shortcomings of OpenAI. Thus, everytime, OpenAI improves itself, it ends up killing a bunch of other platforms.
For instance, with the launch of GPT-4 Turbo, which has an improved context windows from 32k tokens to 128k tokens, Anthropic seems to be losing its relevance. Undoubtedly, GPT-4 is the most advanced large language model (LLM) out there.
AIM spoke to various companies and realised that when it comes to LLM, their first preference is GPT-4. Now, in the case of Claude, Anthropic’s AI assistance, it had the advantage of a context window, supporting 100k tokens. Now, with GPT-4 Turbo, that advantage has vanished. Companies, which preferred GPT-4 over others, will quickly shift to their favourite LLM.
Another killer invention by OpenAI was GPTs that allow anyone to easily build their own GPT without the need of coding. With GPTs, users will have the capability to customise ChatGPT according to their specific needs, making it more supportive in their daily activities, tasks at work or home. They can then share these personalised versions with others.
Moreover, after users craft their personalised GPT, they have the option to share them publicly on the soon-to-be-launched GPT Store, slated for release later this month. Think of the GPT Store as akin to the Google Play Store or Apple's App Store, offering a platform for sharing and accessing a variety of GPT creations.
Once in the store, GPTs become easily searchable and have the potential to ascend the leaderboards. Looking ahead, developers will also have the opportunity to earn income based on the usage metrics of their GPT, creating a financial incentive tied to the popularity of their creations.
This particular invention may sound the death knell for hundreds of companies, which were providing various services built on GPT-4, like DocGPT that allows users to ask any question, from the PDF documents. There are many XGPTs (companies based on GPT-4), which are going to be redundant after this.
Open sourcing act
OpenAI, once known for closed practices, is taking steps towards open source. At DevDay, they unveiled 'large-v3,' an open-source automatic speech recognition model, Whisper, with plans for an API release soon.
Whisper, available on GitHub under a permissive licence, excels in transcribing diverse content and is hailed as a top-notch tool. Its unique timestamp feature makes it ideal for use as subtitles on platforms like YouTube. The model, designed for researchers, segments audio into 30-second clips, leveraging an encoder and decoder for accurate text prediction. Originally intended for integration with ChatGPT, OpenAI opted for a direct public release. Notably, Whisper is currently aimed at researchers, not end users.
The Consistency Decoder, a replacement for Stable Diffusion VAE decoder, was also open-sourced, a notable move for improved stability. However, it's worth noting that the decoder currently supports versions v1 and v2, not SDXL, despite the mention of compatibility with "1.0+."
OpenAI chose to open-source Whisper large-v3 with the goal of providing a foundation for creating practical applications and advancing research in robust speech processing. The AI tool was refined through extensive training on a dataset comprising 680,000 hours of carefully supervised data from the internet. Notably, one-third of the dataset is derived from non-English sources.
AI Forum for India. Our Discord Community for AI Ecosystem, In collaboration with NVIDIA. Join Today >>
TOP STORIES OF THE WEEK >>
Musk’s Open Source Dream
In a recent podcast with Lex Fridman, Elon Musk voiced a strong preference for open-source AI, citing a bias towards openness. Musk's xAI unveiled Grok, a chatbot developed in just four months and currently running on 8,000 NVIDIA A100 GPUs. However, Sam Altman took a dig, creating Grok with a single prompt on GPT Builder.
Musk underscored that OpenAI's foundation was rooted in discussions with Larry Page, emphasising the divergence in views on AI safety. Recounting the recruitment battle for Ilya Sutskever, Musk acknowledged him as the linchpin of OpenAI and revealed his pivotal role in recruiting and funding, lamenting the shift towards closed-source practices.
Read the full story here.
LangChain Knows How to Survive
Despite facing challenges from OpenAI's constant updates, LangChain remains resilient, offering timely responses. OpenAI's recent API releases, especially in replacing traditional models like RAG, pose potential challenges for LangChain in the AI-driven application space. However, users on Reddit express optimism, highlighting LangChain's unique strengths, control, and transparency.
Some users emphasise its flexibility and abstraction layer, making it adaptable to various language models and vector stores. LangChain, despite criticism for design and documentation, stands out for its adaptability and a wide range of services. The recent releases, including LangSmith and LangServe, showcase LangChain's commitment to innovation and staying ahead of developments in large language models.
Read the full story here.
The Tiger’s Tale
Genpact CEO Tiger Tyagarajan, set to retire, leaves behind a legacy of remarkable growth, guiding the company to $4.3 billion in annual revenue in 2022. A proponent of digital transformation beyond technology adoption, Tyagarajan prioritized business reimagination through technology, data, and AI.
His leadership fostered diversity, inclusion, and a focus on ESG goals, aligning with the global shift toward virtual business models and sustainability. Recognized for his unique perspective, Tyagarajan's successor, BK Kalra, takes the helm in February 2024. Tyagarajan remains on Genpact's board post-retirement, with industry leaders expressing gratitude for his 12-year tenure.
Read the full story here.
Recommended by LinkedIn
Harish Sivaramakrishnan: Orchestrating Creativity at CRED
Harish Sivaramakrishnan, Chief of Design at CRED and celebrated Carnatic music singer, discusses the evolution of CRED's design philosophy, highlighting the latest Charcoal iteration's emphasis on clarity and engagement. The minimalist and intuitive interfaces align with the balance between artistic expression and practicality.
Sivaramakrishnan, with a background in chemical engineering, sees parallels between music and design, emphasizing creativity and innovation. CRED's commitment to open-source, exemplified by the NeoPOP framework, aims to enrich the global design community, reflecting the company's dedication to community growth and positive contributions.
Read the full story here.
AIM UPCOMING EVENTS >>
Join AIM at AI Forum for India's exclusive online workshop!
"Responsible Generative AI: Unveiling Risks, Challenges & Best Practices"
📅 Date: Nov 16, 2023
🕔 Time: 5:00 PM IST
Dive deep with Monica Kothari into the ethical landscape of GenAI. Discover how to balance innovation with responsibility.
Get ready for our in-person Meetup in Bangalore - DevPalooza: A Journey into the LLM World! This exclusive event is designed to bring together AI enthusiasts and professionals for a day of knowledge-sharing, exploration, and networking.
Date: December 2nd, 2023
Time: 9:30 AM to 3:30 PM
Location: Analytics India Magazine, Bengaluru
Join Us at India's Biggest AI Conference for Developers - MLDS 2024!
🗓️ February 1 to 2, 2024
📍 NIMHANS Convention Center, Bangalore
The 6th Edition of the Machine Learning Developers Summit (MLDS) is coming to Bangalore, offering the definitive gathering for India's vibrant ML community.
🎟️ Secure Your Spot! Buy Tickets Now! (Early Bird Passes to expire next week)
AIM SHOTS >>