Your business is worth more than 6X yearly earnings

Options
2»

Comments

  • k.l.
    k.l. Posts: 109
    Options
    Thanks for getting this back on track.
  • John Hudson
    John Hudson Posts: 3,034
    Options
    Aside from all the potential risks of AI, a primary issue in any use of the technology is the problem of constrainability. The notion of a ‘robotic intern’ implies something self-contained and working within constraints specifically dictated by human instruction. It is possible to create an air-gapped machine learning tool that operates only on specific training materials provided and only performs tasks specified by human controllers, but that is not the general model of machine learning being applied in today’s most popular AI. To me, such a model seems exactly what I would want from a robotic intern in font making because, as explained in other threads, I would want to train the AI tool to design type and make fonts that way I do so, not the way other people do, but such a model will be viewed by many as unattractive due to the limits it places on the potential of the technology. An AI that has all the Internet as its training material is much more powerful, but also much riskier, since it is not only learning what you want it to learn, but learning things that you may not realise it is learning, including communication protocols.

    So when talking about the usefulness and risks of AI, and when likening it to existing non-AI tools such as automation scripts and algorithms, it is important to consider how constrained or unconstrained is the operating environment of the AI.
  • PabloImpallari
    PabloImpallari Posts: 784
    edited May 2023
    Options
    Haha.. what you want is to indoctrinate your own AI to your own taste... which is totally ok and valid. I have been doing the same.

    You do so by "captioning" your training data. Its a long and tedious process full of thorns, but the results is the sweet nectar in the middle of the rose.

    Once the AI has learned the meaning of the words you have used in your training data captioning, and start to react to the presence or absence of those words in your prompts, you can start to control it as your servant.

    First the AI will have very little to none reaction, but as you keep increasing the training time you will see slow progress... repeat and repeat until results are satisfactory.

    Good idea to save model checkpoints from time to time... so in case you ruin it, you dont have to start all over again.

    One strategy I have been using is to train small models, one each specific for each specific time.... as opposed to train a super huge do-it-all genius monster model. For me is easier, also I don't know enough about AI internals to even aim at the task... for example I still don't know how to "correct" the AI when she learns something incorrect. On those escenarios I need to reset and start over.. but I know that are other better options, including golas and rewards, model merging and many more advanced techniques that may be out of my league.

    In the process of teaching the AI you will learn a lot about you very own tastes too... since many things that one does instinctively without thinking too much about it... will be subject to deep thinking analysis, and you may end up changing your own mind on many regards.
  • MrEchs
    MrEchs Posts: 35
    Options
    Thanks John for the cogent and balanced reply. I appreciate it.

    I certainly didn't intend to say that anyone ought to sell sooner rather than later! So far, the primary factors for most have been lifestyle and financial needs. Different kinds of transactions come with different opportunities for how you'll continue your career. I think it's wise to consider these first. 

    Overall, more parties acquiring type IP portends more parties marketing it. It's hard to see how that does not lead to more revenue opportunities industry-wide, even for those who do not sell. 

    I would expect competitive forces to influence buyers as well as sellers. Buyers will compete with each other to acquire IP. In time this could address some of the issues you mentioned. 

    I know it's a tough business and I have infinite respect for everyone making a go of it. To me it seems there are more ways for designers to earn a living than there were 15 years ago. Some of them are terrible! I hope more choices are better than fewer, and I am hoping that they'll influence each other along the way.

    --Matthew
    Business Letters LLC
    Coaching for Creative People
    http://bizlet.org

  • Nadine Chahine
    Options
    As I mentioned in my ATypI talk, there are now at least 2 entities now active in acquiring foundries: Monotype and The Type Founders. This creates a very interesting dynamic and could potentially explain why Monotype is asking for a 3 months notice period in case a foundry plans to sell to someone else (as mentioned here). How they can ask a foundry to breaks its NDA and disclose business negotiations before they are concluded is beyond my comprehension...

    When it comes to selling and calculating the worth of a foundry, one has to look at its revenue streams and expenses. If its profit is coming from custom work, then its valuation will likely be less attractive because that type of income requires that personnel are kept on board and existing customer relations are maintained. If the buyer keeps the foundry employees, not a problem. If the buyer prefers to keep only the font IP, then that foundry is less attractive.

    If on the other hand, the profit is coming from predictable font licensing, then that is more attractive to a company interested in growing its portfolio.

    Someone earlier asked about the royalties that Monotype takes on fonts. Those are kept by Monotype (as is normal) and do not appear on the balance sheet of the foundry. They were never income for that foundry, and that money was never in the bank account of the foundry. So, not part of the equation.

    One last note, as part of the ILT Academy, we will be offering a seminar on mergers and acquisitions as there's been a lot of questions around that topic. I'll post details later.