Negative Capability

• We talk about how important it is to have the emotional intelligence to hold different, sometimes opposing, ideas at the same time when you’re running a startup.
• When to schedule funding rounds?
• Are you ready to potentially lose more money in the 2024 IPOs? I know I am!
• More AI stuff like running your own local large language models and exploring new prompting methods because, well, are you even alive if you aren’t thrilled about everything happening in this field?

Negative capability, a term coined by John Keats, is the ability to keep two or more ideas that contradict each other, without becoming uncomfortable.

For example: as an entrepreneur, you’re told to forget about trade-offs, but you’re always making them. Take Noa Ganot, who started Infinify. She says creating a product is like building your dream house. First, you dream up the perfect version, no holding back. This approach can really make your end product better. Sure, you’ve got to think about budget, time, and resources at your disposal, but if you start cutting corners too soon, you might kill some great ideas. So, you end up in this weird spot where you’ve got this awesome dream, but it might take ages to even get close to it. Meanwhile, you’ve got to keep going, balancing what you need now against what you’ll need down the road.

You could call that logical negative capability. There’s emotional negative capability too. Let me give you another example. As a founder, you’re bombarded with doubts about your idea from almost everyone you talk to, yet you must march on on as if those doubts were never voiced. On top of that, you have to convince some key people, like early investors and employees, that your idea will, in fact, succeed. Plus, because only the paranoid survive, you should always be on the lookout for any gaps in your strategy. Handling all this demands a great deal of ability to juggle conflicting emotions and viewpoints in your mind.

Most people shy away from that kind of mental juggling. They prefer things simple and clear-cut. If that weren’t the case, everyone would probably be a fan of poetry for its ambiguity, much like Keats was.

Understanding negative capability reveals how much of the conventional corporate world is designed to sidestep it. Career paths are often laid out so people don’t have to think too deeply about their future, following a set progression typical for their role. Feedback gets boiled down to an average of peer opinions, leaving only one clear path to follow. Every superposition of potentially confusing states is collapsed into a single measurement used to quickly and efficiently align team members to row in the same direction. Whole work cultures are built this way.

Being an entrepreneur is about choosing a different route. Without a boss, there’s no one to simplify the uncertainties for you. Sure, you might think the market can guide you, and to some extent, it does. But the market only reaches a clear verdict after a while. This can take years, especially for startups. Think of it like this: the market’s decision usually becomes clear during a major event, like when your company is sold or goes public. That could be 5-10 years down the line. Until then, you’re navigating largely on your own.

Some people I know not only avoid dealing with that level of uncertainty—they just can’t handle it. Their mindset leans towards stability and predictability. And that’s perfectly fine. It works well for most people and is usually very efficient. However, when they try to take on tasks that demand negative capability for long periods, they often end up feeling depressed and burned out. At that point, their productivity takes a hit.

So assess your own capacity for negative capability and choose your path accordingly.

When to schedule funding rounds?

The toughest year for startups with venture capital backing in the past decade is nearing its end. Could this lead to more deals? Carta data from 2019 onwards shows that December often sees a surge in startup deals.

In contrast, January and February tend to be slower for equity fundraising, a trend that’s somewhat mirrored in pre-seed SAFEs and convertible notes, though less markedly. The second quarter is generally the steadiest for deals, with a dip in the third quarter due to summer. Things pick up again in the fourth quarter, leading to a flurry of agreements signed in December.

This pattern is useful for early-stage founders planning their funding rounds. Commit it to memory and leverage it to your benefit!

Potential future IPOs

Keep an eye on these fintech firms potentially going public in 2024, listed randomly:

• Stripe
• Klarna
• Plaid
• Revolut
• Brex
• Deel
• Gusto
• Carta

Clearly, this isn’t investment guidance. I mean, seriously, you shouldn’t take financial advice from me. My interest in IPOs peaked during the 2020 bull market, and so far, I’ve lost money on every one of them. This experience actually persuaded me to stick to early-stage projects 😅

How durable is OpenAI’s lead?

It’s been a year since the launch of GPT-3.5, and the LLMs from major tech companies are still struggling to match its capabilities. Google’s Bard is falling short, while Amazon’s Q and X’s Grok are roughly on par. Meta’s Llama 2 is also similar, and Apple hasn’t even introduced a comparable product.

However, two smaller labs, Anthropic and Inflection, have developed models that surpass GPT-3.5… but not GPT-4. Currently, no one seems to match GPT-4’s level, with the possible exception of Google’s rumored Gemini—who knows. This represents a notably enduring lead for GPT-4. The upcoming year promises to be fascinating, as we wait to see if any lab can close this gap.

Running your own local LLM

This might get a bit technical, so please bear with me. Mozilla’s innovation team, along with Justine Tunney, has just launched llamafile. It might currently be the best method for running open-source LLMs on your personal computer. A llamafile is a comprehensive, multi-gigabyte file that includes both the model weights of an LLM and the necessary code to operate that model. In some instances, it even encompasses a complete local server with a web interface for interaction—similar to having your own local version of ChatGPT.

Let me guide you on starting with LLaVA 1.5. This is a large multimodal model that handles text and image inputs like GPT-4 Vision, fine-tuned on Llama 2. I have personally tried this method on an M1 Mac:

• Download the model from a repository located on Hugging Face: curl -LO https://huggingface.co/jartine/llava-v1.5-7B-GGUF/resolve/main/llamafile-server-0.1-llava-v1.5-7b-q4
• Turn the binary into an executable by running the following command in a terminal: chmod 755 llamafile-server-0.1-llava-v1.5-7b-q4
• Launch the executable you created, and it will initiate a web server on port 8080: ./llamafile-server-0.1-llava-v1.5-7b-q4
• Go to http://127.0.0.1:8080 in your browser to begin interacting with the model.

You might wonder why you should do this. The LLM you’ll operate may not be as advanced as GPT-4, but it’s completely local. You can run it without an internet connection. This means if you can’t use ChatGPT at work due to security or privacy issues, this version could be a viable alternative because it’s entirely private. However, if your employer’s concerns are related to copyright, you might still need to get approval before using it.