Agree, BM25 honestly does an amazing job on its own sometimes, especially if content is technical.
We use it in combination with semantic but sometimes turn off the semantic part to see what happens and are surprised with the robustness of the results.
This would work less well for cross-language or less technical content, however. It's great for acronyms, company or industry specific terms, project names, people, technical phrases, and so on.
Everything I tried previously had very disappointing results. I was trying to get rid of Azure's DocumentIntelligence, which is kind of expensive at scale. The models could often output a portion of a table, but it was nearly impossible to get them to produce a structured output of a large table on a single page; they'd often insert "...rest of table follows" and similar terminations, regardless of different kinds of prompting.
Maybe incremental processing of chunks of the table would have worked, with subsequent stitching, but if Gemini can just process it that would be pretty good.
You don't need to wait for Google. Their Jan 21 checkpoint for their fast reasoning model is available on AIStudio. It shows full reasoning traces. It's very good, much faster than R1, and although they haven't released pricing, based on flash it's going to be quite cheap.
Sure, their 01-21 reasoning model is really good, but there's no pricing for it!
I care mostly about batching in Vertex AI, which is 17-30x times cheaper than competition (whether you use prompt caching or not) while allowing for audio, video, and arbitrary document filetype inputs; unfortunately Gemini 1.5 Pro/Flash have remained the two so-called "stable" options that are available there. I can appreciate Google's experimental models for all I can, but I cannot take them seriously until they allow me to have my sweet, sweet batches.
It's not clear that writing poetry is a bad use case. Reasoning models seem to actually do pretty well with creative writing and poetry. Deepseek's R1, for example, has much better poem structure than the underlying V3, and writers are saying R1 was the first model where they actually felt like it was a useful writing companion. R1 seems to think at length about word choice, correcting structure, pentameter, and so on.
Ok, that makes some sense. I guess I was thinking more about the creative and abstract nature of poetry, the free flowing kind, not so much about rigid structures of meter and rhyme.
Indeed. I would assume that a reasoning model would do far better at things like actually maintaining meter or rhyme scheme, something that models (even with good attention mechanisms) generally do very poorly at.
You can just use it and see for yourself. It's quite good.
I do believe they were honest in the paper, but the $5.5m training cost (for v3) is defined in a limited way: only the GPU cost at $2/hr for the one training run they did that resulted in the final V3 model. Headcount, overhead, experimentation, and R&D trial costs are not included. The paper had something like 150 people on it, so obviously total costs are quite a bit higher than the limited scope cost they disclosed, and also they didn't disclose R1 costs.
Still, though, the model is quite good, there are quite a few independent benchmarks showing it's pretty competent, and it definitely passes the smell test in actual use (unlike many of Microsoft's models which seem to be gamed on benchmarks).
Sorta. If you subdivide an area into more pixels you get lower dynamic range for each pixel, but you can average across multiple pixels to gain back the some of the dynamic range.
It’s not about data, it’s customer segmentation. Frequent customers are more price sensitive, and are willing to use the app to get all the discounts, while occasional customers will not, so they can capture both the more price sensitive part of the market while getting higher margins from occasional buyers.
As someone who spent many years segmenting customers and generating personalized marketing offers -- McDonald's is awful at this. I was a 2-3x/monthly customer (USA based) for years (even more frequent a decade ago, but I'm talking about since the app), ordering the exact same core items every time (except during breakfast).
When they began "value meals" last summer (which don't include their flagship items) they also removed the best deals from the app, the ones that did include Big Mac, QPC, 10-nuggets. I've placed one non-breakfast order in 6-8 months, whenever they started this.
I'm just one person, but if a customer declines from an expected 15-20 visits over a half-year period to 1, and you don't adjust your offer algorithm (and you're the biggest restaurant company in the world so no lack of resources), something is seriously wrong.
Whenever this happens to me I keep wondering how much I am of the A/B data test where I'm in the "less important group". Is it possible that their changes engaged (or profited from) the more active (daily/weekly customers) by making your situation worse?
Perhaps. Let's assume that the value meals is a massive hit and they are collecting far more revenue from customers who like it, than they are losing from people like me.
That's the whole point of data analytics and personalized marketing - even if the value meal works for most people they can still go back to sending me the offers and promotions I responded to previously, in an attempt to reverse my recent decline in spend/visitation. The app makes it possible to send individualized offers. There shouldn't be an entire "B" group where they just say, oh well.
They used to have great deals on the app in Germany. I used to go to McDonald's all the time. The deals suck now, and now I only go if I'm really craving a McMuffin Bacon & Egg.
They've captured the user base with the money that corporate was pumping into the app deals, and are in the process of enshittifying it by transferring the value to themselves instead of the users.
This can work in a lot of industries - I am skeptical fast food is one of them. Switching costs are low, alternates are plentiful, and collecting information (reviewing deals/prices across companies) is relatively easy.
If McDonald's enshittifies its deals while continuing to raise prices, it's way too easy for loyal customers to go elsewhere. I'm saying this as a huge fan and extremely loyal customer of McDonald's for decades... they are at serious risk of losing people like me. As I stated, I've gone from 15-20 visits to 1 since last June/July, whenever they made the big change.
We've got similar opinions here. I'm just pointing out that the overall experience here feels familiar, and it wasn't until reading this thread that I really put it together.
I agree with you that I'd be surprised if Enshittification works as well here as it does in tech, but maybe since there's an app involved, they just think they can get away with it. Who knows.
I don't understand why people even think about selling $10/mo licenses.
Just go sell $2-4k/mo SaaS to a medium sized traditional industry that has ancient, crummy software with a target customer employee size of 50 - 1,000 people.
Customer problem discovery is easier, you get reputation network effects within the vertical, you gain trust because you learn the industry lingo, you can actually solve peoples' problems because you're not balancing needs against unrelated industries, they're willing to pay, and once you get 20 customers you're doing pretty OK.
> Just go sell $2-4k/mo SaaS to a medium sized traditional industry that has ancient, crummy software with a target customer employee size of 50 - 1,000 people.
I’ve successfully done exactly this.
Except in order to get there, we had to spend 3-4 years selling cheap $29/mo self-serve plans at volume to iron out all of the kinks in the product and UX. Only after that were we able to go upmarket selling $40-90k/yr contracts.
It all boils down to the fact that customers paying $20k/yr or more have extremely high expectations across the board. It’s extremely difficult to just pop into a completely new market with a brand new product untested by anyone, and expect large companies to trust and buy from you.
True and what I find funny is that the same companies are often willing to pay a lot of money to consultants that will deliver a shitty streamlit POC. Like they will literally pay consultants the equivalent of 4 years of new SaaS cost to get a POC instead of using a product with rough edges and grow with it. Expectations can be really dumb and confusing.
I'd pay to buy a bad house and then spend money to improve it, rather than paying a full rent for the same bad house and hoping that the landlord uses that money to improve it.
I don't get the comparison. are you comparing a SaaS that is doing 70% of the job and a shitty streamlit POC that no more than 3 people can use in parallel ??
They’re saying that they would rather pay a consultant because they would have control over fixing things that don’t work the way they want. Compared to the alternative of buying a SaaS product that is “rough around the edges” where you’re not able to directly control whether and how the product is improved over time.
To be clear: It usually makes no sense to develop an internal solution when there is an estabilished offering on the market. If there isn't, embracing an rough and incomplete Saas solution from a startup may work, but:
- You have no control on how the solution will evolve
- You don't own the solution (an internal solution can became a product on itself)
- You don't have any competitive advange (the money you pay are used to develop a product that benefit your competitors as well as you).
Yeah exactly. You can't just roll out of bed one morning and grustle out a stable enterprise application. Also:
> customers paying $20k/yr or more have extremely high expectations across the board.
They definitely do, in certain ways. But also a lot of the software companies use sucks shit. But this part of the other comment cracked me up.
> a medium sized traditional industry that has ancient, crummy software
This is actually how to do it but the thing they're leaving out is that you have to already have knowledge of the specific industry and its players, their niches, and the software they use within them. You have to know where standards are high, and where the current options are bad, and how to market that improvement to the companies that are juuuust flexible enough to switch to your product while being conservative enough not to dump you if exactly one VP leaves.
Once you have all that in place, sure, you can "just" sell your software to them lol.
>This is actually how to do it but the thing they're leaving out is that you have to already have knowledge of the specific industry and its players, their niches, and the software they use within them.
In most cases, this requires at least one former employee who used the crummy software. Most likely this person would be the one to identify the problem and propose an alternative solution.
Information like this is extremely valuable. It forms the core of the biz opportunity. Casual comments often overlook this.
Builders doing their own thing may not overlap with employees using office horror software 9-5.
Boiled down into 1 sentence: Create a company that solves a problem you’ve personally experienced.
(If you feel like you haven’t experienced any problems worth creating a company to fix, consider putting yourself in situations where you’ll be exposed to different/new problem sets)
Yes, get industry experience! You have to be pretty special to be just a software developer with no domain knowledge who can somehow make something more valuable than the bazillion other just-a-software-developers out there. I sell fairly high priced software to companies and I'm doing fine. Though I'm kind of lucky in how I got into it and by now most of my domain knowledge has come from my previous customers asking for things.
We experienced something similar - we were able to wrangle a $30k/year contract as one of our first contracts, but the expectations around the software were and continue to be incredibly high - we sort of wish he had had a few $100/mo customers first to iron out the UX, and just learn more about how to work with a midmarket/enterprise buyer.
Anyway we survived but we understand why companies move upmarket over time now haha.
Hell, I went straight for 20k/mo (average) with my SaaS.
The key is providing a good value proposition. My software would save my customers more than they would spend on it yearly, while outsourcing $50-300k/year in salary worth of work that previously had to be done in internally (usually incorrectly).
You always need to solve an actual business problem that people are willing to pay to fix.
Would you be willing to share insights and help with mentoring in private. I'm going this path (building SaaS) and would appreciate someone with experience to share to discuss topic.
There's gotta be a name for this fallacy. "This enterprise software sucks and is extremely expensive, so we will build something that is actually good and become billionaires!"
This is the graveyard of engineer driven startups in the enterprise space, who make the mistake of thinking that software quality matters. The fact is it is not a top priority, if it is even a priority at all.
In the enterprise space, if you're not a Fortune 500 company, all that matters is 'relationships', by which I mean sleazy sales processes (or actual nepotistic friends-and-family relationships).
Joel Spolsky wrote a great essay once about why there is no software priced between $1000 and $100,000. Because above $1000 you need a sales team that takes purchasing managers to strip clubs and play golf with CEOs, and that shit adds up. On the plus side, your software can be dogshit, like the stuff you work with every day.
Only engineer and product oriented types think that quality has any actual value. Unfortunately it does not, even less so if your only USP is "quality". If your products only differentiator is that it's technically better than the competitor it really needs to be 100x better in order to sway anyone's interest.
If your maximum work/effort you can put into any project is 100% which you have to allocate into fragments, the product person will put 80% into product development and 20% to marketing but a sales person will put 80% into marketing and 20% into development and unfortunately the latter will product much better results. All it needs to be is "good enough" but reaching to your target market with the right product really is the key.
The unfortunate side effect is that quality is best achieved by having the PR team write it out and put it into a snappy slogan on your product website. Done. Quality achieved.
Edit: And the above applies to your typical non-safety critical Saas, Enterprise, B2B, B2C software. And thinking back to for example the Boeing crashes it likely also applies to the the "safety critical" software/systems as well...
You can get very far with just sales and marketing, but if the quality isn't there, reputationally, that's going to catch up with you over time, same if you have dodgy business practices (i.e.: predatory pricing practices for existing customers, impossible cancellations etc.).
The world has changed. Today you can sell software for $5000 or $10000 without any sales process. Not always of course, you have to find the right niche.
There is a whole non-corporate world of SMBs that employ 30 or 100 people total. Those businesses have no VPs or corporate cards, but they still have problems and are willing to pay good money for solutions. Family businesses in any industry, law offices, marketing agencies, boutique hedge funds/pe, etc.
Corporate approval for a 10k purchase doesn't need to be complicated either. As a software vendor it helps to have a web page/email with all the relevant info intended to be forwarded to accounts payable.
>Just go sell $2-4k/mo SaaS to a medium sized traditional industry that has ancient, crummy software with a target customer employee size of 50 - 1,000 people.
The guy is so out of touch it's like a cartoon. Like, right, why did they think of that?! Just create something that's worth several thousands a month instead of a small value product!!! And get all the big companies who can actually afford that kind of thing to buy your product!!! How hard can that be?
The only times I've seen anything remotely like that work, it's a situation like this: Two engineers work together at company A, become friends and gain experience in the domain. Both move on, progress in their careers and work other companies in the same domain. They meet for coffee and person 1 says: "hey remember that problem we had at A? I see it again at B", person 2: "me too and I've been working on a side project to fix it, you should totally buy the solution from me". And voilà. Could a kid right out of school build the same thing? Sure. Would it have ever succeeded? No. They lack the domain experience and the social connections to do it.
I didn't say it's easy, and it's probably not the right fit for you. But it's at least a plan with a good end-game, which a lot of $10/mo online only business models simply cannot turn into something viable.
50-1,000 person companies aren't actually that big. You can get to know the decision makers and it's not like navigating 20,000 person corporate mazes (also possible if you have the stomach for it).
It does seriously benefit from industry knowledge though. The reason so many people think it's a ridiculous pipedream is because they have no industry experience outside of tech, and that's a big problem -- a solvable one, but most people will never do what it takes to solve it and instead say that it's not possible.
It takes a lot of willingness to do non-tech things though. You can't just set up an online marketing funnel and watch the cash come in. It's relationship driven. The advantage of going after a vertical is that the relationships start feeding each other -- you get social proof within the niche industry. And you don't compete with VC-backed software companies as much because they much prefer horizontal solutions (which can get bigger and achieve bigger exits) versus vertical solutions.
Yes. But it's not for everyone, especially not for a lot of engineers.
Many of these buyers feel more comfortable meeting you in-person, so going to trade shows, writing articles for niche trade publications, and sometimes doing site visits are all very helpful. So is cold calling.
Sales cycles are often slow. You have to get used to customers taking months to decide, having lots of meetings with different stakeholders, and to get good at telling the difference between those who need a lot of meetings but will likely close versus those who will never do anything. It's a very different game than selling subs online, which I know little about.
The product does need to be decent but I disagree with others that it has to be amazing. Most enterprise software is horrible. They will be delighted if you actually solve their problems (rather than the problems horizontally-oriented software vendors want their customers to have), even if it has some rough edges, as long as you're very customer-focused, responsive, and available.
But it also tends to be durable revenue, because you get tightly integrated into their operations if you do it right.
The one point where gp is wrong is you really can't sell $4k things; $10k min and $25k is better. eg a good to great salesperson closes 25% of deals, so you have the cost of 4 mqls per one close, and then the salesperson gets 20%ish of the deal.
Just gotta find that price point within discretionary spending where a manager can OK the cost without having to get a VP to sign off on it. It's definitely way higher than $10/mo.
Norway doesn't care. It is a country with a reputation for good governance and northern-European economic strength. But economically, it is a country that is largely a gas station: like a democratic Russia with more competent governance.
Over half its economy is based on oil and mining. It has failed to develop meaningful economic diversification, and, because it has wisely banked so much of the proceeds of its oil (over US$300k per capita), there's not a lot of pressure to adapt.
Norway will not be a center of innovation anytime soon, except in oil-related fields. Eventually, as oil gets replaced as a source of energy, they may feel more pressure to change. But for now, they suffer from a more sophisticated version of the resource curse.
Sweden is an interesting counterexample, which has a lower GDP per capita but a much more diversified economy. Sweden abolished a wealth tax they used to have almost 20 years ago.
Over half of the value of exports is oil and mining, but oil production, mining and quarrying directly employ only about 23,000 people, compared to 190,000 in manufacturing for example.
Because such a large portion of the oil proceeds are banked, it also distorts the rest of the economy far less than it otherwise would. Unless you live in very specific parts of Norway, you can go your entire life with hardly any exposure to the oil industry directly, or secondary interactions with major suppliers to the oil industry.
I recognize your point but am wary of it being supported with direct job creation numbers, as that doesn’t reflect the scale of secondary industries (ie. indirect jobs created by oil & mining may be a larger number than direct jobs created) and also, job numbers don’t reflect the scale of the companies creating the jobs.
There's now doubt it'd have secondary effects, which is why I didn't try to make a precise comparison. Shutting down the oil industry would certainly make a difference. Hence the focus on shuffling so much of it into the oil fund (the Norwegian sovereign wealth fund owns ~1.5% of all listed shares globally at this point) to compensate, but there's no realistic scenario where it'd account for a large proportion of jobs.
In terms of value, yes, it'd have a much greater impact, which is why they're being taxed at rates that even for Norway are absolutely extreme, and why the oil fund does not invest inside Norway, and why such a small proportion of the oil funds investment returns are spent each year - it's been a very intentional government policy for decades to both prepare for a future without the oil income and to reduce the effect it has.
We use it in combination with semantic but sometimes turn off the semantic part to see what happens and are surprised with the robustness of the results.
This would work less well for cross-language or less technical content, however. It's great for acronyms, company or industry specific terms, project names, people, technical phrases, and so on.
reply