The Beacon: Hard Science Fiction by Brandon Morris (red white and royal blue hardcover TXT) 📕

confirmation code from his credit card company, and received a welcome message.

There was no mention of a new language model. He hoped he hadn’t fallen for a scam. He logged in with the more recent data.

“Vg unf ab frys.”

Again the meaningless translation appeared. But this time, when he clicked on the icon with the dumbbell, something happened. A gear wheel turned, and a red text appeared on the screen.

“Too little data to build language model.”

Well, that was no problem. He entered the whole text.

“vg unf ab frys—vg vf rirelguvat naq abguvat. Vg unf ab punenpgre—vg rawblf yvtug naq funqr; vg yvirf va thfgb, or vg sbhy be snve, uvtu be ybj, evpu be cbbe, zrna be ryringrq. Vg unf nf zhpu qryvtug va pbaprvivat na Vntb nf na Vzbtra. Jung fubpxf gur iveghbhf cuvybfbcure qryvtugf gur punzryrba cbrg. Vg qbrf ab unez sebz vgf eryvfu bs gur qnex fvqr bs guvatf nal zber guna sebz vgf gnfgr sbe gur oevtug bar, orpnhfr gurl obgu raq va fcrphyngvba. N cbrg vf gur zbfg hacbrgvpny bs nal guvat va rkvfgrapr orpnhfr ur unf ab vqragvgl—ur vf pbagvahnyyl vasbe[zvat] naq svyyvat fbzr bgure obql. Gur fha, gur zbba, gur frn naq zra naq jbzra jub ner perngherf bs vzchyfr ner cbrgvpny naq unir nobhg gurz na hapunatrnoyr nggevohgr—gur cbrg unf abar, ab vqragvgl. Ur vf pregnvayl gur zbfg hacbrgvpny bs nyy tbq’f perngherf.”

Was that enough for the algorithm? It had to be enough, because that was all he had. This time, the gear wheel turned longer. Hope rose. But could it be so easy to decode a radio transmission that, he hardly dared to think, had come from aliens?

“Too little data to build language model.”

Of course, it wouldn’t be that simple. Crap. So what now?

He wasn’t one to give up. First, he tried again with a second version of the text, in which he had shifted the beginning by one bit. It was silly because it didn’t change the amount of text, but it didn’t cost anything to try. The gear wheel turned for a while.

“Too little data to build language model.”

He had a hunch, so he squelched frustration and anger. He could still increase the amount of text without additional input. He only had to try it with a 4- or 3-bit encoding. He recalculated the text. In the first variant, it became 20 percent longer; in the second, another 25 percent. Now would it be enough? He copied the 4-bit text into the window and pressed the ‘Start’ button.

“Too little data to build language model.”

So the 4-bit encoding was not enough. He tried the even more extended version. The gear wheel turned longer than ever before. Finally, a green line appeared.

“Creating new language model.”

Peter rocked back and forth on his chair. He mustn’t get too excited too soon. This was machine learning, not understanding. The algorithm was trying its hardest to impose the structures of a language on the data. How many languages might the software know? What was certain was that these were only human languages. Was alien communication subject to similar laws?

The system spit out a translation:

An impertinent Internet believes the evident repentance. Moreover, it means the profitable initiative. The critical systems confirm the diabolic paradigms to the stringent partnership. They entrust to the situation the suspect experiment. Nevertheless, they obey the transparent enthusiasm and taste the final technology. The successive schemes tell the diverse weaknesses a posthumous construct. For this, they fall in after the artificial positions.

Peter laughed. The text reminded him of some of his principal’s speeches. But there was no sense in it. The result just proved that the program could squeeze something out of some data that read remotely like meaningful text.

But at least he now knew how much input the software needed to start working on it in the first place. He needed to double the amount of text at the very least. Or, better, quintuple it if he wanted to be sure.

He needed Thomas’s help again. There was no other way.

March 21, 2026 – Passau

“Vg unf ab frys. That sounds funny,” Thomas said.

“I don’t know what’s funny about it,” Peter said.

“You’ve always been a buzzkill. No wonder your wife... Sorry.”

And Thomas had always been an empathy-less asshole. No wonder he lived alone. But now Peter needed him.

“Can we maybe have a serious talk?”

“Look, if you click on the speaker icon, the program even tries to pronounce ‘vg unf ab frys’ for you.”

“I know. So, then what? Can you help me?”

Thomas audibly drew in a breath and then exhaled it loudly. Peter couldn’t see his face because he was just sharing the screen content with him, but Thomas made it pretty clear he was annoyed. Peter could understand him. From a purely scientific point of view, this was all very thin.

“It’s the weekend, but I’ll fit it in somehow. I really don’t know if that’s going to help you, though.”

“Yes, it will. You already helped me a lot. But, I need five times the amount of data to be able to obtain something meaningful.”

“Peter, you’re getting lost in this. I almost feel like a co-dependent getting you your dope. You’d be better off using your time to get your wife back. I know what a loyal dog you are. I don’t mean that disparagingly. I really don’t. I couldn’t do it, being with the same person for so long, but your path has always been clear to you. Remember when we used to get drunk together?”

“And you’d tell me which girl you’d most like to hook up with. But then she’d turn you down. I remember. Girl after girl.”

“Yes, because I was a penniless student. We should really get together again sometime. Why don’t you come by here when you’ve got your affairs in order? You can bring Franziska, too.”

“First, I need your help to get my affairs in order.”

“Do you want me to call Franziska for you, as an old friend?”

“No, Thomas, I’m talking about the data.