Month: January 2023

Microsoft finds itself more than ever cuddling up to OpenAI’s shoulder. After Redmond’s billion-dollar investment in the startup and the future integration of ChatGPT in Bing, this is not a big surprise either. The technology company is now experimenting with the linguistic model to integrate it with Word, PowerPoint, Excel y Outlook.

Microsoft Office 2021 DOWNLOAD

According to The Information, Microsoft has already incorporated it into Word to improve the autocomplete function and generate text through simple instructions. However, it is undeniable that the company still has a long way to go. Engineers and developers are working hard to create tools for writing e-mails and documents using artificial intelligence. However, it is not clear that Microsoft will launch these new features for the time being.

Thanks to ChatGPT’s integration with the Office suite, we could see our workflow become even more streamlined. For example, Outlook could be able to compose entire emails through simple prompts from us. However, Microsoft faces a number of challenges that it must address, such as ChatGPT’s tendency to present erroneous information in some cases. This would make it very difficult to create documents, for example.

Another major hurdle is privacy. Microsoft will have to customize the generative models used to make them work without compromising user data. For the time being, the Xbox owner is working on the development of models based on GPT-3 and GPT-4 that preserve the privacy of those who use them.

The truth is that the technology developed by OpenAI opens up many possibilities. Its models could be used to transcribe Microsoft Teams meetings, transfer data to PowerPoint presentations and, in turn, integrate images generated by DALL-E 2. However, for the moment it is all just wishful thinking: we do not know when we will see this type of functionality in Microsoft applications and it seems that we will have to wait a while yet.

Artificial intelligence requires human intervention. As clever or resourceful as this class of technology may seem, there’s an inherent issue with AI that isn’t likely to be solved anytime soon. The issue is that artificial intelligence uses up information far faster than we humans can produce. And, let’s not misunderstand the situation here; we create original content, AI simply spins, reworks, infers, and finds information. 

This issue plagues generation AI more than anything else. These are tools like large language models such as ChatGPT, and image generation software such as Dall-E. Language models like ChatGPT require a significant amount of training on all sorts of information. In fact, the utility itself makes this quite clear if your prompt includes queries on events that happened after 2021. You’ll get a neat little paragraph in which ChatGPT asserts that it was trained on specific information and can only answer questions or provide context concerning information and events prior to 2021. 

However, the vast stores of information that these systems are trained on is what makes them seem so knowledgeable and highly educated. There’s been a massive movement to train these systems on more and more data so that their perceived abilities and breadth of knowledge improve even further. But, unfortunately, we are not running out of data. 

A recent research paper published by researchers from Epoch provides further information about this issue, and according to the paper, we could run out of training data for artificial intelligence by as early as 2026. It’s not only the researchers at Epoch who feel this way, though. Teven Le Scao, a researcher at Hugging Face, an AI company, has also asserted that he fears we may soon run out of appropriate data with which to train further artificial intelligence initiatives. 

Researchers at Epoch say that the problem stems primarily from the way that data is sorted. When data is selected to be used in AI training, it gets sorted into two piles; high-quality, and low-quality. However, there’s apparently a very blurry line between the two piles. Nonetheless, there’s a difference between that which usually falls cleanly into the category of high-quality data, and that which does not.

High-quality data is usually written by professional writers, while social media posts, 4chan rants, and general commentary most often qualify as low-quality data. The issue is that there is far more low-quality data than high, and AI researchers prefer to use high-quality data when training artificial intelligence because of the intended result of the initiative. People want to interact with high quality content. 

Some researchers assert that it may be time to reassess what qualifies as low quality data. However, this carries the significant risk of ending up with chat bots riddled with baised human opinions and thoughts that are then circulated to whoever may be using the utility at the time. We’ll have to wait for researchers to battle this one out, but it certainly seems like AI’s longevity could be called into question soon. We may be looking at just another tech fad.

Even though there are fears that our current interpretation of AI technology is due to hit a ceiling around 2026, the industry is still evolving with every new initiative released. The latest initiative is Microsoft’s Vall-E, and it can replicate and mimic human speech in a matter of seconds. 

The impressive part here is that this isn’t the first time that a company has attempted to create an AI that can mimic human speech. However, previous attempts have routinely proven how difficult and time-consuming such an enterprise is. The core issue seems to be that it takes far too long for these systems to learn individual voices, not to mention the vocal intricacies that each person instinctively employs. 

Microsoft has done something truly remarkable here. Vall-E has astonished almost the entire tech community with the way that Vall-E is able to replicate and mimic human speech in an incredibly short time. In fact, it only takes a few seconds. When we say a few seconds, we truly do mean no more than a few seconds. On average, Vall-E only needs around 3 seconds of speech to be able to replicate someone’s voice, intonation, and general vocal idiosyncrasies. These few seconds of speech required to train Vall-E in replicating a human voice has gone down as the smallest sample size that the industry has ever seen. 

If you’re interested, researchers at Cornell University recently released a paper on how Vall-E works. The paper also breaks down all the differences between Vall-E and other text-to-speech synthesizers. 

Here is an excerpt from the paper that’ll impress science and technology wards. ‘Large-scale data crawled from the Internet cannot meet the requirement, and always lead to performance degradation. Because the training data is relatively small, current TTS systems still suffer from poor generalization. Speaker similarity and speech naturalness decline dramatically for unseen speakers in the zero-shot scenario.’

‘VALL-E significantly outperforms the state-of-the-art zero-shot TTS system [Casanova et al., 2022b] in terms of speech naturalness and speaker similarity, with +0.12 comparative mean option score (CMOS) and +0.93 similarity mean option score (SMOS) improvement on LibriSpeech. VALL-E also beats the baseline on VCTK with +0.11 SMOS and +0.23 CMOS improvements.’
In simple terms, very smart researchers at Cornell University have found a way to do something that was thought to be relatively impossible. As GHacks recently reported, Apple Books has released an AI tool that can turn any book into an audiobook.

However, the utility has faced harsh criticism over the way the tool sounds. I listened to the utility at work and found it pleasing, but others have not been as kind. The release of a tool like Vall-E, however, could possibly revolutionize the audiobook industry and intensify the good work that Apple has initiated.