jacobr1

29 ms·

391.
by jacobr1 3y ago
> I could see copy-cat services being more of an issue in the late stage of a product though? When everyone knows lots about how to stand it up and use it? The concern isn't random small companies. The concern is the big cloud provi
392.
by jacobr1 3y ago
Isn't it already relatively cheap to run? Training is costly, but there are examples of running LLaMA on your laptop. It doesn't seem like it will take decades to commoditize this stuff ... it is just the cutting edge might keep m
393.
by jacobr1 4y ago
> An actual professorship at UT even if it were not possible to be tenured would surely be preferable to those options I'd think. And also it includes a gamble that texas will reinstate tenure.
394.
by jacobr1 4y ago
Sure, but I also could just copy it to millions of people. But in this context we are talking about artificially limiting usage. So we could do something like limit sharing to once a day, or month or whatever. No sharing or transfer seems t
395.
by jacobr1 4y ago
It would be reasonable to allow for the exclusive sharing or transfer of an ebook. Why can't I resell the single-copy (and thus forfeit my own use) or lend it to a friend when done (and again I would be unable to use it when it was len
396.
by jacobr1 4y ago
Your odds are much better with 3 companies, that one might succeed. Especially if you are joining late-stage startups. But at the same time, your equity package will be smaller, so the exit won't be as impactful.
397.
by jacobr1 4y ago
I'm not sure. Methods like Chinchilla and Quantization have been able to reduce compute by more than an order of magnitude. There might very well be a few more levels of optimizations within the same statistical paradigm.
398.
by jacobr1 4y ago
It already is relatively trivial to fine-tune generative models for various use cases. Which implies huge gains to be had with targeted applications not just for niche players but also OpenAI and others to either build that fine-tuning into
399.
by jacobr1 4y ago
Yes. One of the frontiers of current research seems to be multi-modal models.
400.
by jacobr1 4y ago
It would be better at an intermediate cert level. Those are generally more aligned to an organization's boundaries.
401.
by jacobr1 4y ago
Correct. Those lower-desk-utilization offices still have current leases.
402.
by jacobr1 4y ago
Looks like they have the integration half of the equation. But also they need to plugin the summarization/synthesis utility to fully realize the value. There have been enterprise search apps in the past and they usually have failed bec
403.
by jacobr1 4y ago
It is also a lesson for the cost-cutting that is popular right now. Sure, trim fat and waste - but R&D and and test projects in new areas are important for future potential markets. Keep some of that alive even it currently isn't p
404.
by jacobr1 4y ago
We need someone to build this that indexes the myriad of corporate data hidden in various docs and saas systems.
405.
by jacobr1 4y ago
I think the relevant distinction is concentration. In other US cities with relatively high crime - the crime is concentrated in certain neighborhood and is heavily related to gang activity. So you provided you can avoid certain areas, you g
406.
by jacobr1 4y ago
We used sqlite for a queue at a prior startup - enabling the WAL was critical to maintain disk throughput.
407.
by jacobr1 4y ago
The basic thesis, "More and more major businesses and industries are being run on software and delivered as online services" does still seem to be directionally correct. It isn't the full story, atoms still matter, but the bi
408.
by jacobr1 4y ago
I do this semi-periodically now. At the end of the week I close all tabs, I archive everything in my inbox, mark all items in my slack as read ...
409.
by jacobr1 4y ago
What amazes me when I talk to people that work in "non-tech" companies is how common boilerplate work is actually out there. Man-millennia of effort to move data from one system to another, or generate a report, or trigger somethi
410.
by jacobr1 4y ago
Yep, this is coming. But it is unclear how interactive will be in practice and what the feedback mechanisms will look like.
411.
by jacobr1 4y ago
Yes - but that would be as a extension/integration with their browser asset - not a random unrelated project.
412.
by jacobr1 4y ago
For anybody comfortable with the compiling the single-file binary ... it doesn't gain you anything. For a class of "power-end-users" it provides a mechanism to build sandboxed apps on a multi-tenant system. I think the spirit
413.
by jacobr1 4y ago
The point isn't the government services are currently high quality. The point is that the current environment foster low quality service. In some cases voting in politicians that drove top-down management changes would help. And that w
414.
by jacobr1 4y ago
Logarithmic scaling factors for your management layers. Below 50, you basically have a flat org. Even with nominal mangers, everyone is effectively on the same team. 50-150 you need an actual management layer, but these managers are all on
415.
by jacobr1 4y ago
It might also be wrong, yes. I have plenty of code licensed under very permissive licenses that still requires attribution. It is an open question of how much the AI system is a "derived" work in a specific, technical sense. And i
416.
by jacobr1 4y ago
The prior deal was based on royalties for use. Adobe pays you 33% percent of anything they make. It is consignment. So if someone uses a specific photo for $20, you get paid $6.60, no money us paid upfront. So what should adobe pay you for
417.
by jacobr1 4y ago
> As an aside, the name Bard is so low-effort and predictable, and a bit rubbish in my personal opinion. I would suggest the opposite. Choosing a good name is hard. You need to avoid naming conflicts, and ideally you have something short
418.
by jacobr1 4y ago
Yeah but other systems are more open. Undoubtedly it is about legal risk management - yet goog should have gotten its lawyers to draft whatever global TOS they needed to protect themselves up front. This has enough visibility and they had t
419.
by jacobr1 4y ago
Some of these app are too niche, but maybe not all of them. There have been many attempts at a "unified enterprise search app" for example, but they all have generally failed. Maybe customer indexing + GPT + form fact (slackbot, g
420.
by jacobr1 4y ago
One approach I've been thinking about is focusing on enhancing the Behaviour-Driven Development (BDD) workflows. I was was really into cucumber and its tooling ecosystem a decade or so ago. One of the challenges is maintaining these &q

More ›