The doozy AI summer continues

The last time I wrote, the AI arms race was having a crazy summer week: Llama 3.1 launched alongside OpenAI's search, and Cohere raised a punchy $500M. The pace hasn't let up — large-model press releases continue to be rampant, regulation is posturing at the extremities, and the market is asking, out loud now, whether this is a bubble.

The EU released its 144-page AI Act, which is mainly (a) posturing at the extremities of risk management around development for hostile uses, and (b) virtue-signalling to broader centre-left Europeans and Sam Altman sceptics. Intel, meanwhile, laid off 15,000 people, citing decreasing revenues tied to various failures in its AI initiatives.

Hours into August, Google's Gemini 1.5 Pro overtook OpenAI's GPT-4o in a significant community benchmark — GPT-4o scored 1,286 to Gemini's previous 1,261. For the past year or so Anthropic and OpenAI had been the dominating contenders; no longer. AI firm Galileo also released a hallucination index focused on RAG, putting Anthropic's Claude 3.5 Sonnet ahead of Gemini 1.5 Flash.

Is the bubble bursting?

Google has advanced in AI-robots, developing droids that beat humans at ping pong. Amusing cyborg anecdotes aside, the core sentiment across markets has been one of volatility, with recent easing — the VIX calming after its four-year high is probably just a momentary, much-needed mid-August sigh of relief in a high-octane market.

News that Apple side-stepped Nvidia in favour of chips designed by Google for its AI infrastructure didn't help Google amid wider market woes. Investors cited caution about the rate of return across AI and chip technologies, and in some cases were all-out freaked.

On chips — taking NVIDIA as the case in point — commentators including the respected hedge fund Elliott Management have described a plausible bubble. The basic thesis: unless exceptional, sustained execution is matched with sustained high-growth demand, NVIDIA will struggle to realise the trillions in returns it needs to recoup value for its enormous market cap. It shed 20% of its share price between July and August, only to win most of it back in the last week.

I suspect this is more of a slowing, or a frothing, than a bursting. The market needs to figure out where and how to apply generative AI for pragmatic, valuable use cases at scale. As for the foundational hardware side, NVIDIA has a mountain to climb in a hugely complex, multi-faceted economic model — chip design and supply chain are almost continually on the science-innovation event horizon, and that's before we add geopolitical and natural-resource constraints. Heavy, man.

Scale challenges

This is really a case of two deeply intertwined industry economic models needing to become efficient at an unprecedented pace: the success of chip manufacturers in hitting delivery targets, and the pace at which large companies deploy AI at scale to create sustained demand.

The underlying GPU layer is incredibly complicated and could slow things down properly. Nvidia reported delays to its next AI chip due to a design flaw affecting the substrate layer exclusive to the Blackwell family; its workaround, the B200A, leans on an older, simpler — but less future-proofed — CoWoS-S technology instead of the problematic CoWoS-L.

Zuckerberg and Huang say every company will have an AI. Of course they're bullish, but they're also sound-biting for us mere mortals. The truer statement is that every business will have an ever-evolving, complex web of interacting AI.

Scale is still hampered by limited clarity on where AI should be deployed beyond established uses such as translation, where companies are already racing to maximise value (see DeepL). Part of the strategic-level inertia is really about how AI can be deployed accurately and safely. As firms figure out how to deploy responsibly, they'll also figure out the targeted areas to deploy. We're early in this contextual evolution, but the signs point to complex webs of LLMs combined with multiple neural networks and proprietary data sources. So yes, every company will have an AI — but a static, single LLM? Likely not. These models will become increasingly context-aware and integrated, each company running a living, evolving language model.

Fresh money

Fear not — there's still plenty of capital being dished out at eye-watering, industry-shifting levels. This remains the best time in history for growth, industry, innovation and quality of life. Canva announced its purchase of Leonardo.Ai and its generative image model, accelerating its move to become a genuine AI-native leader. Groq's value reached $2.8bn after a reported $640M round to take on Nvidia, and named Meta's chief AI scientist Yann LeCun its newest technical adviser.

Necklaces and AI

And finally, on an uncynical note: the genuinely human — rather than dystopian — story behind Mark Zuckerberg's new necklace. Worth the 48 seconds.