The Leading Edge #13

Hello, Leading Edge Enthusiasts!

Lucky number 13! Welcome to another edition of your go-to weekly digest where we peel back the layers of the tech world's latest and not-always-greatest. This week, we're hopping right into the thick of it with a tale of two AI devices that might just have bitten off more than they could chew. ????

Oh, and let's not forget the mysterious "GPT2-Chatbot" that appeared and vanished like a crypto bro's millions. Was it a sneaky glimpse of GPT-5? Or just Al's machine learning cousin pulling a prank?

Buckle up, folks, because the AI roller coaster is only getting wilder!

AI Devices: Another One Bites the Dust

Fresh off the spectacular debacle that was the Humane AI pin, it seems the Rabbit R1 also hopped into the market a bit too prematurely. The tech community has been buzzing, and not in a good way.

Let’s dive into what’s been said.

MKBHD called the device "barely reviewable"

And he’s not the only critic.

The Verge gave it a horrible review here, and the reviewer said "All the Rabbit R1 does right now is make me tear my hair out."

Engagdet review is here, and they called it "A $199 AI toy that fails at almost everything".

So, why all the disdain?

Three main reasons stand out:

• Underdevelopment: The device feels more like a prototype—clunky, slow, and not ready for prime time.
• Terrible Battery Life: Engagdet review notes that R1 would burn through it's battery in just 8 hours while doing "absolutely nothing"!
• Your Smartphone is Better: Like the Humane AI Pin, your existing smartphone does everything better!

The so called Large Action Model is also missing in action. This indeed the whole reason why they went viral in the first place. A fancy name and a good twist on the LLMs.

Unfortunately this marketing trick does not hold up in execution. It had just four integrations:

• Spotify
• Uber
• DoorDash
• MidJourney (Why? What makes this a priority?)

But here’s the real kicker: it doesn’t do them well. Many reviewers identify this as providing incorrect responses, being highly onconsistent, taking too long and that in general their smartphones do a much better job in terms of the tasks.

So, why does the Rabbit R1 exist as a standalone device? Why not just a sleek, integrated app? Android Police did some investigations and they realized most of it could be just an Android App.

So what did we learn from these multiple disasters?

My biggest problem is that unfinished devices like Humane AI pin, and Rabbit R1 are cashing on the AI Hype chain and using that to attract VC funding - eventually to let everyone down. So you may invent some cool marketing terms like LAM - and you may go viral - but you just can't deliver.

The challenge of outperforming the smartphone is monumental—it requires more than a handful of engineers and a half-baked device. However, not all is bleak in the realm of AI devices. Meta’s Smart Glasses, for instance, offer genuinely new use cases and complement smartphones, though even they struggle with low adoption rates.

The takeaway? For now, it might be best to stick to your smartphone and wait for giants like Google or Apple to integrate more robust generative AI features.

The True Opensourcing of LLMs: A Game Changer in AI

With the recent unveiling of Llama3, a seismic shift occurred in the world of large language models (LLMs): open source models are now giving commercial giants a run for their money.

The LMSYS leaderboard is buzzing with excitement, placing Llama-3-70B in the top 5, rubbing shoulders with industry titans like GPT-4, Claude 3 Opus, and Gemini 1.5 Pro. What's even more impressive is that Llama3 isn't just competing—it's outperforming established commercial models like Gemini Pro and Claude 3 Sonnet.

That's no small feat.

Let's delve a bit deeper into what this means for you and me. Imagine running a GPT-4 class model right from your home office. You'd need a robust setup—think a graphics card along the lines of the RTX 4090 and running a 4-bit or smaller quantized model — like this one.

But the results?

They'd be remarkably good.

Now, for those tinkering on their local PCs, I’d suggest opting for the 8B model for snappier responses.

Meanwhile, amidst all this buzz, Mistral seems to have gotten lost in the shuffle. Dropped down to position 13 on the leaderboard, this commercial model's parent company is now seeking a whopping $500M funding, eyeing a $5B valuation. But one has to wonder: with no clear discussion of their earnings, how do they justify these figures? They'd need to be pulling in at least $200M in annual recurring revenue to make sense of a 25x multiple. Their party tricky seems to be that they are the Europe's only AI messiah.

As for OpenAI, they are now like the employee doing barely enough not to get fired. After Clause 3 Opus breifly took the crown, they made a model update on 9th April to get back to the top. Clearly they are not interested in doing anything more for GPT4, and would rather work on GPT 5. Everyone's now eyeing the horizon for GPT-5, especially after a hinted release on April 22 came and went without fanfare.

Adding to the intrigue, "GPT2-Chatbot" appeared mysteriously in the LMSYS arena last week and disappeared just as mysteriously. Not quite revolutionary enough to herald the arrival of GPT-5, but enough to stir some lively debates.

You can see a 17 minute review from Mathew Berman on the "GPT2-Chatbot" below:

So, what's your take on these developments? Was it GPT 4.5 or perhaps Llama3 400B stirring the pot?

I'm eager to hear your thoughts and conspiracy theories.

Until next week...

We're still growing - this is the 13th edition! Your feedback stays as crucial as ever. Hit reply and let me know what you think! Want to see a specific topic covered next week? Don't be a stranger, share your ideas!

And of course, if you find this newsletter valuable, spread the knowledge! Share it with your network and help us grow. ????

See you next week!

#artificialintelligence? #generativeai #leadership #ai #productdevelopment #startups