kozikow

Father claims Google's AI product fuelled son's delusional spiral

I got into quite a lot of rabbit holes with AI. Most of them were "productive", some of them were not.

80% it will talk you out of delusions or obviously dumb ideas. 20% of the time it will reinforce them

I did lose 30-40kg about 2 years ago on Ozempic.

I don't count calories. I went off Ozempic (now Mounjaro) and I gain weight at about 0.5-1kg a month.

As I am resistance (gym) training, significant % of that ends up being muscle mass rather than fat.

So I end up taking Mounjaro for about 1-2 months every 3-4 months, approximately 33% of the time being "on".

Funnily, I end up with bulk/cut periods without doing them explicitly. This ends up working well for growing muscles.

Notice all people in the story are women. I guess pairing GLP-likes with bodybuilding works quite well for men. As times goes on, I end up needing mounjaro less due to my increased muscle mass.

The H-1B program was already broken by the lottery. This new fee just solidifies the L-1 visa as the real high-skilled pipeline. More L-1 visas are already approved annually than new H-1Bs, and this policy only widens that gap.

In addition to L1, O1 is also often gamed. $100K for H1B is mostly "posturing" at this point, as voters don't know about other options.

Ads inside LLMs (e.g. pay $ to boost your product in LLM recommendation) is going to be a big thing.

My guess is that Google/OpenAI are eyeing each other - whoever does this first.

Why would that work? It's a proven business model. Example: I use LLMs for product research (e.g. which washing machine to buy). Retailer pays if link to their website is included in the results. Don't want to pay? Then redirect the user to buy it on Walmart instead of Amazon.

And realize the thing on your head adds absolutely nothing to the interaction.

There are some nice effects - simulating sword fighting, shooting, etc.

It's just benefits still outweigh the cost. Getting to "good enough" for most people is just not possible in short and midterm.

Mindless optimization of basic "attention grab" metric is why the whole internet feels like a slots machine. Be it reddit, Facebook, YouTube, any google result

Thankfully this won't happen with LLMs, as compute is too expensive so execs can't just take an easy way out of optimizing for number of questions asked

More to that - at this point, it feels to me, that arenas are getting too focused on fitting user preferences rather than actual model quality.

In reality I prefer different model, for different things, and quite often it's because model X is tuned to return more of my preference - e.g. Gemini tends to be usually the best in non-english, chatgpt works better for me personally for health questions, ...

I am big fan of "cost monitoring".

In my previous company I had a good setup for costs monitoring - including release to release comparisons, drill downs, statistics, etc.

After each release I looked at this data. It saved a lot of $, by simple fixes like "why we are calling this API twice?".

It also quite some issues that weren't strictly customer related, but weren't apparent from other type of data (you will always have some "unknown unknowns" in your monitoring, and costs data seem to be pretty wide net to catch some of those)

Chatgpt content is getting pasted all over the web. Now, for anyone crawling the web, it's hard to not include some chatgpt outputs.

So even if you put some "watermarks" in your AI generation, it's plausible defense to find publicly posted content with those watermarks.

Maybe it's explained in the article, but I can't access it, as it's paywalled.

the harder it is for me to use these tools in a way that doesn’t feel like too much blind faith (even if it works!)

I tend to ask multiple models and if they all give me roughly the same answer, then it's probably right.

Cersei Lannister: Power is power.

Knowledge is a necessary, but not sufficient component of power

Or in other words observability is a necessary, but not sufficient component of optimization.

Or other way around - in bigcorp (or in startup) choosing what to work on have much bigger impact than the work you do.

On very low level it's up to your manager. As time goes, even as IC you have a lot of agency. It's not just company selection, team selection, but also which part of the project you are working on and how you are approaching solving it.

Of course "if everyone does this, who will fix the bugs". However, the quickest promoted people I've seen are the people who were excellent at politics-izing (and sometimes foresight) the best work assigned to them.

I used Ozempic for couple months. I lost 25kg over 6 months (120kg -> 95kg).

I gained muscle, as I started weightlifting (modified 5x5 program 3-4 times a week) and was supplementing with protein isolate (about 50g a day).

My subjective feeling is that even if "Ozempic makes you lose muscle faster than the same caloric deficit without it" is true, this effect is very small.

Vast majority of muscle loss comes from no resistance exercise, low protein, much faster weight loss than possible "naturally".

"The theory behind these models so aggressively lags the engineering"

The problem is that 99% of theories are hard to scale.

I am not an expert, as I work adjacent to this field, but I see the inverse - dumbing down theory to increase parallelism/scalability.

Story of two engineers in a team I worked with.

P is what most people would consider 10x engineer (but not this article). Can get anything done few times quicker than anyone else. It's like 10 junior engineers stacked in one person. But it often would be unmaintainable mess.

M is a lot like article describes. Understands what needs to be done and creates good technical designs. Often unf*s mess created by P. Delivers business value. Do not write a lot of code.

It's funny that depending on whom you ask, M or P would be the 10x engineer and other would be the bad engineer. Real 10x engineer can wear the M or P hat depending on circumstances.

A $1.3 billion revenue company being too tight to pay this after all, even on their 2nd chance, is so short-sighted it's absurd.

I'll give an "another side" perspective. My company was much smaller. Out of 10+ "I found a vulnerability" emails I got last year, all were something like mass-produced emails generated based on an automated vulnerability scanning tool.

Investigating all of those for "is it really an issue" is more work than it seems. For many companies looking to improve security, there are higher ROI things to do than investigating all of those emails.

Ah, I see someone who doesn't commute by car

What do you think about product by Google - Google Distributed Cloud Air-Gapped?

Although the name is "air gapped" - it does not have to be if client doesn't want air gapping. It's "buy commodity hardware from vendors HP, we will give you software and training to manage it".

Much leaner stack than whole GCP/AWS/Azure, but deployed "on-prem" with "cloud-like" experience.

Not a Googler, but gpt4o has been a declaration of war by OpenAI and Google is shutting down things no longer considered to have long-term strategic value.

Flying car essentially exists - it's called helicopter.

It's just uneconomical to do actual flying cars due to costs and fuel inefficiency. Flying cars will happen when if we have breakthroughs in battery technology and potentially energy generation.

Any production deployment would probably do pre-fetching of potential initial dialogues as some kind of background job.

Nono, let stealing the internet first, generate synthetic training data, and then outlaw it to ensure no one else catches up.

What's your feeling those types of lawsuits will land? Probably many people are also facing product decisions today keeping in mind the potential precedence law in the future.

I kind of feel that in general with AI regulation, the USA will have to go all-in somehow. EU will delve into future obscurity, but this technology is too valuable and China is breathing behind's USA neck, so I expect law to continue to be lax around IP of data used to train AI models.

For start, day trading is effectively gambling.

Not unless you managed to get hired as a (quant) trader with a performance fee. Privatized gains, shared losses. Of course, it's very hard to get hired for such a role.

I'm currently doing BI evaluation for improvement of the current toolkit, so I might use this comment section for some BI experts' advice.

We mostly use Looker Studio + Metabase currently.

Investigating Tableau, Looker, PowerBI, Sisense, Omni currently. PowerBI is great, but BigQuery integration seems to be lagging. Is there a way to get such a tool without paying $50K a year? Everything so far is only a "nice to have" upgrade over our mostly free toolkit. We also care about good geospatial visualizations and embedding reports for users without requiring a login.

Where GPT really improved for me subjectively is long context - at least since launch. Such ranking can't compare it.

Also I felt I keep getting more personalized results, i.e. models are somehow biased towards user. I heard they plan it, I don't know it's launched, but I feel it.

And there's also fine-tuning in the other direction - my brain got used to ways of interacting with GPT. Same as with Google, I just somehow subconsciously know how to write prompts that get me what I want.

All of those variables they described are a contention point in other coffee-making methods like espresso as well.

And many of those options are "correct", but depend on your preference - do you like more concentrated coffee? Do you like it less bitter? Do you want more caffeine? Etc.

I think many founders have some buried deep worry like this.

You end up fundraising, preparing forecasts and budgets for investors for the next X years, attending conferences, working with marketing on PR, dealing with unglamorous problems that are not part of anyone's job but need to get done and someone else doing it would impact morale too much. It's lucky if you find 1-2 days a week for "deep work" given constant context switching.

Even if you work 2 shifts you can't sometimes match the level of being necessary for day-to-day operations as "regular" employees. But you own orders of magnitude more equity.

I'm more interested how those geospatial visualization were done. Do you know a probable library used for it?

I think what should be mentioned in the article is bridge rounds. I know various startups that have done it. In my feeling, at least half of the 38% would be explained by VCs raising bridge rounds for their existing portfolio companies.

Rather than come clean that all your valuations came down by 2-3x (the reality of what valuations are circulating on the market nowadays) just do things like raise convertible note with a cap of last round valuation and don't announce it on pitchbook/crunchbase.

It's not easy for VCs to just pause investing or invest in some other stuff than their "thesis" - some have agreements legally binding on what they would invest in and timeframes for spending the capital.