jamez1
jamez1 on Hacker News
In hindsight, it was obvious LLMs would be good at this. The neural net weights are embeddings for the circuits within, no different from the embeddings that represent textual or visual tokens.
You probably don't need Astra capability to produce this, just lots of training data on the same NN showing prompts and their activations.
....Databricks has no revenue relative to its valuation either
To be a good engineer you had to be stubborn in the early days of your career, you had to have tenacity and persistence to understand and solve the problems you saw. Unfortunately that filter selects for people who are prone to going too far into the detail, who get a satisfaction from conquering complexity rather than avoiding it. Given the lack of big picture thinking, they readily fall into herd-like behavior. Given the ego payoff for the complexity, it is very difficult for them to accept criticisms about their process or lack of perception.
Skill loss works both ways. You might miss out on forming early skills in using llms effectively, and end up playing catch up in 3-5 years from now if LLMs mark all the skills you hold to be void.
It is also likely LLMs will change programming languages, we will probably move to more formal, type safe languages that the LLM can work with better. You might be good at your language but find the world shifts to a new one that everyone has to use LLMs to be effective for.
Ontological frameworks have been around for many decades. They have had limited success because it is very difficult to represent knowledge in an object oriented approach. An example is Ologs:
https://en.wikipedia.org/wiki/Olog
It would be interesting to know what you see is different about your approach that would lead to more adoption?
Interestingly, no acknowledgement has been made on the stock exchange under the ticker (CLBT), despite the two days of public knowledge
It is logical to assume self driving cars start off in low risk areas, prove their concept, and gradually take on more and more use cases. Things like retirement villages and so on, where this incremental approach has already had success. Skepticism is easy when thinking about the grand scale but this incrementalism will surely win out in the end.
Given the complexity of the setting, there is far more possible causations than the implied by the finders of this correlation.
For example, perhaps better funded hospitals have more females on staff, and that drives the clinical outcomes.
Or maybe this relationship was one of thousands of possible relationships, as it is clear they have examined as much, and it's just appeared statistically significant by chance.
Hornsdale power reserve derives most of it's revenue from selling ancillary services, it does not provide energy. Hornsdale's success is not foreshadowing a battery revolution unfortunately.
The reason nuclear costs so much is because we've stopped building plants. There's no industry anymore, let alone economies of scale. It's a fallacious argument to say it costs too much. If we went all in on nuclear to tackle climate change the costs would quickly collapse. After all, they were built in the 70s and dramatically decarbonized the industry at the time.
No, the enterprise value of the company doesn't change. It was owned by 4b of debt before and is now owned by 5b of equity, and has 1b in cash. It also has an intact business.
Under your logic no company would ever raise equity to pay off debt.
Great example of how difficult statistical concepts are to understand for the layperson, even if that person happens to be a judge or a lawyer.
This would be far better if the currency was used to pay for connectivity, and the router was reimbursed for routing.
Connectivity has a cost, and none of these networks will amount to anything unless the model can incentivize all aspects of the infrastructure necessary
They've made a terribly flawed assumption which drives that 90% number. VE is modeled using a beta binomial model with a fixed prior of an infection rate of 1.5% per year. Obviously this number is going to vary a lot depending on the sample. [1]
By putting a fixed point vs the appropriate wide ranging distribution as the prior, the model scores the results far higher than otherwise.
This is a non-result and it was wildly inappropriate to publish this as proof of any finding.
[1] https://pfe-pfizercom-d8-prod.s3.amazonaws.com/2020-11/C4591... Page 15
And funnily enough, 50 years ago the only feasible solution to move off fossil fuels was nuclear. Which was then campaigned against by the green lobby, not the fossil fuel lobby. The irony is that the only reason we haven't de-carbonized is because the same people who are yelling about climate change decided to stop it.
Uh this is just wrong, institutions use algorithms to trade, <10 contracts isn't confirmation that it's retail money. The algorithms purpose is to hide the flow, and they mimic order size at the very least.
BayesDB provides a toolset, what you've proposed is where you need knowledge of the underlying process. You've shown some initiative here, but I'd really recommend studying what's out there and doing a true contrast of your solution to see the shortfalls.
Sorry, this is ludicrous. No analyst is basing their investment view on the 1 day post-IPO movements. Maybe a few months in, as an ancedote about market sentiment, but 5 years in the stock is traded totally differently, by different holders in a basically a different world. When funds make investment decisions we collate a whole bunch of random facts, some relevant, some just general interest. 1 day returns off the IPO can be filed under general interest, just like the gossip magazine articles written about the founders. At best it's attached to the fame of the stock, not the other way around - Facebook's pop is memorable not because of the pop but because of the stock.
Maybe because you're from the sell side you're hearing what you want to, it often puzzles me some of the views that you guys have on the market, the role you have just doesn't actually matter that much to the market. A stock is a business, not an investment banking product. The investment banking actions over the life of the business are out-sized, but they also quickly fade to be indiscernible - Facebook would still be the same company today and worth the same amount today even if it was backdoor listed on the pink sheets at first.
This looks like a fairly low power monte-carlo system. You just store samples, and the inference is sampling that sample set? That's just bootstrapping and has been far more explored in a much more extensive way by random forests and so forth.
You've shoe-horned this logic that usually belongs in linear algebra world into a database table form, there's some initiative there but this has also been heavily explored in the academic field under the topic of Probabilistic Databases. BayesDB is a full implementation of what you've just described, with a much deeper inference engine utilizing joint distributions rather than just distributions that exactly match the sample.
So effectively, you've added a set of ML/statistics scripts to the query engine? But the query engine is otherwise still relational based?
I'd like to learn a bit more about your architecture/process and why it creates value over the standard ML toolkit.
It makes sense philosophically to increase capability in the database to handle uncertainty and so forth. Databases were built for transactions not analytics and a rethink would likely be fruitful.
Also, do you have any funding?
This is just nonsense, patents don't accrue much of the revenue and the revenue for the entire telecommunications industry globally is probably only a trillion.
Taking the subsidies out and examining the profitability of all the companies on the supply chain tends to add up to a loss. Given there isn't much labor involved, energy costs are a good candidate for the lack of profitability.
If the input energy costs more than the money the turbine makes, there's a good chance that the turbine doesn't make enough energy to produce itself.
Everyone in quantitative finance has built a back-test that would make them billions on paper. It is highly likely there is over-fitting issues, errors in calculations of returns etc.
You should share the list of trades, rough high level idea of what you're making money from etc.
If you have something, trade it with capital. Nobody seriously buys signals for good reasons.
The population is collapsing due to the decline in fertility, the only growth is from increased life expectancy in the elderly and sub-Saharan africa high fertility rates, both of which are slated to end in the next few decades.
There is no physical phenomenon that causes the log-linear trend, its just a model that happens to fit for a small select period of history. Moore's law isn't an actual law I hope you understand that..
Rising prosperity comes with rising energy consumption. If China didn't export so much it wouldn't be able to import the fuel that it used. Its dramatic rise was definitely not a given.
Renewables are less stable than fossil fuels, that results in too much supply at times, and too little at others. These are facts not FUD.