Happy Wednesday!

In Today’s Email:

• What to expect from Nvidia’s earnings 💸

• Meta’s latest AI model 🚨

• GPT-3.5 Turbo 💨

What to expect from Nvidia’s earnings 💸

How long will Nvidia lead the charge in AI hardware?

It’s common knowledge that Nvidia has been the kingpin of the AI hardware industry. The market for chips powering generative AI products has caused the company’s stock to triple this year. Nvidia is favored by corporations and start-ups due to its software programming platform, CUDA. For over a decade, developers have been building and sharing AI-related software on the platform, which makes it easier for new companies to rapidly build and scale AI applications.

Many analysts estimate the current quarter’s revenue for the company to be near $12.6 billion. Morgan Stanley analyst, Joseph Moore, recently doubled down on his $500 price target for shares. “We expect a very positive outlook, but it’s important to recognize that the exact revenue number in October is less important than the overall trajectory,” he wrote.

Nvidia’s demand is far outweighing supply. The critical question is, for how long? Investors need to see the company showing progress toward fixing supply constraints over the longer term. Nividia reports earnings today after close, hopefully, we’ll see what’s on the horizon for the chip manufacturer.

Meta’s latest AI model 🚨

The tech giant recently released a groundbreaking new AI model.

The AI model coined SeamlessM4T, is a multimodal model that translates speech to text and vice versa. The model can translate speech-to-speech and text-to-speech outputs for 35 languages. Translation for speech-to-text, speech-to-speech, text-to-speech, and text-to-text inputs supports up to 100 languages.

Meta stated that SeamslessM4T is “the first all-in-one multilingual multimodal AI translation and transcription model.” This means the model can translate and transcribe languages at the same time. Meta is also releasing the training dataset powering the model, SeamlessAlign, which has 270,000 hours of speech and text alignments.

Similar to other AI models released by Meta, such as Llama 2 and AudioCraft, SeamlessM4T is publically available to developers and researchers with a research license. Meta’s method of releasing AI models benefits the company tremendously. AI ethicists who rally for transparency of AI systems are happy, whilst developers are simultaneously improving and building the products.

GPT-3.5 Turbo 💨

OpenAI recently released GPT-3.5 Turbo.

The update allows developers to run customized models that perform better for their specific needs at scale. According to OpenAI, “early tests have shown a fine-tuned version of GPT-3.5 Turbo can match, or even outperform, base GPT-4-level capabilities on certain narrow tasks.”

Fine-tuning customers will be able to improve model performance use cases. Improved steerability will make the model better at following instructions, such as always responding in a given language. Reliable output formatting will improve the model’s ability to consistently format responses, which will be especially useful for code completion and API calls. Custom tone will improve the model’s qualitative feel to best fit the needs of the business.

The update will also enable businesses to shorten their prompts while maintaining similar performance. The model can also handle 4k tokens, which is double that of previous fine-tuned models. Support for fine-tuning with function calling and GPT-3.5-turbo-16k as well as fine-tuning for GPT-4 are coming this fall.

That’s it for today, we’ll see you back here Friday!