10 investors discuss AI’s future and what’s beyond the ChatGPT craze

Future Of Artificial Intelligence (AI)
Source: forbes.com

An unusual response was received from one of the investors after Tech Crunch brought up “the rise of AI” in a recent email to them: “The ‘rise of AI’ is a bit of a misnomer.”

Rudina Seseri, a managing partner at Glasswing Ventures, is the investor in question. She argues that deep learning and other advanced technologies have been there for a while, and that all the excitement around them is obscuring the truth that these technologies have been under development for decades. She said that 2010 witnessed the initial enterprise usage.

Nevertheless, there’s no denying that artificial intelligence (AI) is garnering previously unheard-of amounts of interest, and businesses everywhere are busy considering how AI may affect their sector and beyond.

Partner at Earlybird Venture Capital Dr. Andre Retterath believes a number of things are combining to create this momentum. “We are seeing the ideal AI storm, where the three key components that have developed over the last 70 years—access to powerful compute, large-scale datasets, and advanced algorithms—have finally come together,” he stated.

Even still, we couldn’t help but be dubious about the quantity of startups that earlier this year presented a variation of “ChatGPT for X” at Y Combinator’s winter Demo Day. What is the probability that they will continue to exist in a few years?

As this is a trend that businesses cannot afford to ignore, Karin Klein, a founding partner at Bloomberg Beta, believes it is preferable to run the race and risk failing than to wait it out. “The greater risk is to pass up the chance, even if we’ve seen a number of ‘copilots for [insert industry]’ that might not be here in a few years. Now is the moment to start experimenting with AI in your organisation, or else it will lag behind.

What holds true for the typical business also holds true for startups: It would be a mistake to ignore AI completely. However, a startup must also stay ahead of the curve more than the typical business does, and in some AI-related fields, “now” may already be “too late.”

Related Article: ChatGPT Doesn’t Exist Where Real Generative AI Innovations Lies

We surveyed a small number of investors about the future of artificial intelligence (AI), the areas they see the most potential in, the potential development of multilingual LLMs and audio generation, and the value of proprietary data in order to better understand where startups still have a chance and where oligopoly dynamics and first-mover advantages are shaping up.

This is the first of three parts of a study designed to delve deeply into artificial intelligence and the evolution of the field. You will hear from other investors on the many pieces of the AI jigsaw, where companies stand the best chance of succeeding, and where open source could eventually surpass closed source in the following two sections, which will be released shortly.

Tech Crunch also spoke with:

  1. Manish Singhal, founding partner, pi Ventures
  2. Lily Lyman, Chris Gardner, Richard Dulude and Brian Devaney of Underscore VC
  3. Rudina Seseri, founder and managing partner, Glasswing Ventures
  4. Karin Klein, founding partner, Bloomberg Beta
  5. Dr. Andre Retterath, partner, Earlybird Venture Capital
  6. Xavier Lazarus, partner, Elaia
  7. Matt Cohen, managing partner, Ripple Ventures

Pi Ventures’ founding partner, Manish Singhal

In the upcoming years, will the top Gen AI models of today and the firms that created them continue to be at the top?

The environment of LLM applications is one that is constantly evolving. Only a small number of businesses in the application domain will be able to grow from the many that form. We definitely anticipate competition from other parties in the future, particularly in terms of foundation models, for OpenAI. It won’t be simple to remove them, though, as they have a substantial advantage.

Which AI-related businesses, in your opinion, lack the innovation to survive for another five years?

There need to be a major consolidation in the field of applied AI, in my opinion. Applied AI firms that rely on off-the-shelf models may find it difficult to maintain their moats as AI becomes more and more horizontal.

On the other hand, both the infrastructure (tools and platforms) and applied fronts are seeing a significant amount of fundamental innovation. They’re probably going to perform better than the others.

Is becoming open source the most apparent path for entrepreneurs in artificial intelligence?

Depending on the goal of your solution. It’s a viable approach for the infrastructure layer firms, but it might not work for everyone. Depending on the issue being solved, one must decide whether or not to use open source.

Would you like to see more LLMs with training in languages other than English? What other forms of distinction do you anticipate seeing in addition to linguistic differentiation?

Although LLMs are available in other languages as well, English is the language that is utilised the most frequently. LLMs in many languages make perfect sense based on the local use cases.

In addition to language distinction, we anticipate that LLM variations with specialised knowledge in fields like finance, law, and medical will offer more pertinent and accurate information in those areas. In this field, some work has already been done, such as Bloomberg GPT and BioGPT.

When using LLMs in actual production-grade systems, you run the risk of them being hallucinated and irrelevant. To make them more useful right out of the box, I believe a lot of effort will be done in that area.

How likely is it that the way neural networks are now built using the LLM technique will be altered in the next few weeks or months?

It’s not impossible, however it could take more than a few months. Once quantum computing becomes widely used, the field of AI will undergo yet another major transformation.

Are other forms of media, like as generative audio and picture production, underappreciated in comparison to ChatGPT’s hype?

The use of multimodal generative AI is growing. These are necessary to develop for the majority of serious applications, particularly those using text and graphics. Audio is a unique case: considerable research is being done on voice cloning and music auto-generation, both of which have enormous economic potential.

Aside from these, auto-coding is growing in popularity, and creating videos is an intriguing new angle — eventually, we’ll see entirely AI-generated films!

Do you think startups with proprietary data are worth more now than they were before AI became so popular?

Contrary to popular belief, having private data provides you a competitive advantage, but ultimately maintaining confidentiality of your data becomes quite challenging.

Thus, the tech moat is the result of combining data with well crafted algorithms that are productized and optimised for a particular purpose.

If at all, when may artificial general intelligence become a reality?

With some applications, we are approaching human capabilities, but a truly artificial general intelligence is still far off. It could take a very long time to reach there consistently since, in my opinion, the curve is asymptotically steep after a certain point.

A number of technologies, including behavioural science and neurology, may also need to converge for full AGI.

Do you think it matters whether the businesses you invest in participate in lobbying efforts and/or focus groups about artificial intelligence’s future?

Not really. Our businesses focus more on finding solutions to certain issues, and most of the time, lobbying is ineffective. Participating in discussion groups is beneficial since it allows one to monitor how things are progressing.

Glasswing Ventures’ founder and managing partner, Rudina Seseri

In the upcoming years, will the top-performing generation AI models of today and the firms that developed them continue to lead?

Over time, it is probable that the foundation layer model suppliers, namely Alphabet, Microsoft/OpenAI, and Meta, will continue to hold their dominant market positions and operate as an oligopoly. Nonetheless, there are chances for competition in models that provide notable distinction, such as Cohere and other well-funded companies at the fundamental level, with a focus on privacy and trust.

We haven’t invested in the generative AI base layer and probably won’t. There are two likely outcomes for this layer: There is a possibility that the foundation layer may exhibit oligopoly dynamics similar to those observed in the cloud market, meaning that a small number of firms would obtain the majority of the value.

The third argument is that the open source ecosystem provides a major portion of the foundation models. We believe that the greatest potential for venture investors and founders is in the application layer. Businesses may overtake major incumbents in established categories and establish dominance in new ones by providing their clients with observable, quantifiable value.

Our investment approach is specifically targeted at businesses providing foundation model enhancements through value-added technologies.

Considerable value creation remains to be created throughout the entire AI stack, just as it did not stop with cloud computing infrastructure providers. The race against artificial intelligence is far from over.

Which AI-related businesses, in your opinion, lack the innovation to survive for another five years?

Certain AI market niches could not be viable as long-term ventures. The “GPT wrapper” category, which comprises products or solutions based on OpenAI’s GPT technology, is one instance of this. These solutions don’t stand out from the competition and are readily overtaken by features introduced by industry leaders already in place. As a result, they will eventually find it difficult to keep a competitive advantage.

In a similar vein, businesses won’t last if they don’t offer substantial commercial value or don’t address a need in a lucrative, high-value industry. Think about this: Unlike a platform that solves intricate problems for a chief architect and provides unique, high-value advantages, a solution that simplifies a simple activity for an intern will not grow into a major corporation.

Lastly, deployment and acceptance will be difficult for businesses whose products need significant upfront expenditures or do not fit well into existing organisational workflows and structures. This will be a major barrier to producing a meaningful return on investment as it will raise the bar when expensive design upgrades and behaviour adjustments are needed.

Click here to read full article.