Growing up most of us were taught to always say “please” and “thank you”. It’s a necessary to be part of a respectful and polite society. Now the head of OpenAI is asking us to stop – at least when it comes to talking with ChatGPT.
The reason? Too much processing power is apparently wasted on these polite comments. It might seem weird and there have been endless debates on whether we should stop being polite just because we’re working with AI, but that’s not really the point here.
It’s all about power
If two small words are taking up so much processing power, what are the demands of everyday prompts? We type away, amazed at the outputs we receive in seconds, but do we ever stop to think what the systems are built on? As the adoption of AI applications grow, will the broadband structure most companies operate on, be able to support it?
AI demands capacity, speed, and not average processing speed but very high speed. It also demands low latency as most AI applications want real-time processing. Combined, these three requirements place a significantly higher demand on bandwidth capacity and processing capabilities, compared to what companies have been using before. As AI evolves, broadband networks will need to keep up. What could this look like in an AI powered world?
Extending fiber networks
In general, fiber is seen as the backbone for AI infrastructure, because of its ability to support high speed, high-capacity data transfer. It’s not just about extending the cabling footprint though. Switches, servers and network connectors have to be able to match the network capacity otherwise they become bottlenecks that slow everything down.
Optimizing networks
Because AI places such high demands on infrastructure, a key element to expanding and improving networks is having the ability to optimize them for traffic. Managing peak loads will be critical. While the purpose of this is to support AI applications, AI itself has a key role to play in helping optimization.
Network security
AI has the potential to quickly identify vulnerabilities and security risks in networks. It also has the ability to use this knowledge to circumvent them in the hands of bad actors. As networks are systematically extended and upgraded to accommodate AI, approaches to security will need to adapt accordingly.
Scalability
Most companies wanting to run with AI are realizing the step cost associated with it. Even overhauling internal networking infrastructure is a major task and an expensive one. Then there’s the challenge of ensuring scalability – given how rapidly AI applications are advancing, how long will today’s upgrades last? Does this mean more funding needs to be made available to constantly keep up with AI’s demands?
Broadband companies are a key part of the AI revolution, and while it may be fun being in on the action, it’s not going to be easy to keep pace. Bigger, faster, better networks, secure too… will they be able to deliver?
