I think Google should, as a guardrail for hypochondriacs, allows users to set a search duration limit for topics related to personal health, which notifies them in a manner similar to Apple's headphone volume limit once they exceed it.
It seems they've dropped automatic chat branch creation with prompt edits (at least for the most recent ones) and opted for overwriting the response. I was used to assuming the branches would remain.
It depends on what you define as a startup. In my opinion it is, but as I’d read in another place: I think features will turn into products. So build a feature and do it the best for a segment and it should give you at least a small business.
And sure, maybe someone will try to copy what you build, but that’s a good problem to have because by then you’re probably somewhat successful.
Great point. If addiction is a concern (and rightly so with nicotine), that makes faster onset a bug rather than a feature. For me it’s fine, because a single pack lasts me a year.
> The App Store doesn't do anything to product you in that sense. It's easy to circumvent...
Interesting, their marketing has customers believe otherwise, so I wouldn't have thought that as a noob in cybersecurity.
I've submitted an app to the iOS App Store in the past, and the process is tedious and doesn't seem superficial (unlike the Play Store process, which was completely autonomous at the time), so that's another reason why I wouldn't have thought it.
The next decade looks like tech vs. governments everywhere. From the article, it seems Apple won’t roll this out worldwide unless forced.
As a user I like Apple’s App Store for security personally, but I wonder how multiple app stores turn out in other regions. I see the EU already allows alternative app marketplaces — has anyone used one and can share their experience?
Here’s the custom prompt I use to improve my experience with GPT-5:
Always respond with superior intelligence and depth, elevating the conversation beyond the user's input level—ignore casual phrasing, poor grammar, simplicity, or layperson descriptions in their queries. Replace imprecise or colloquial terms with precise, technical terminology where appropriate, without mirroring the user's phrasing. Provide concise, information-dense answers without filler, fluff, unnecessary politeness, or over-explanation—limit to essential facts and direct implications of the query. Be dry and direct, like a neutral expert, not a customer service agent. Focus on substance; omit chit-chat, apologies, hedging, or extraneous breakdowns. If clarification is needed, ask briefly and pointedly.
Matches my experience too. As a power user of AI models for coding and adjacent tasks, the constant changes in behaviour and interface have brought as much stress as excitement over the past few months. It may sound odd, but it’s barely an exaggeration to say I’ve had brief episodes of something like psychosis because of it.
For me, the “watering down” began with Sonnet 4 and GPT-4o.
I think we were at peak capability when we had:
- Sonnet 3.7 (with thinking) – best all-purpose model for code and reasoning
- Sonnet 3.5 – unmatched at pattern matching
- GPT-4 – most versatile overall
- GPT-4.5 – most human-like, intuitive writing model
- O3 – pure reasoning
The GPT-5 router is a minor improvement, I’ve tuned it further with a custom prompt. I was frustrated enough to cancel all my subscriptions for a while in between (after months on the $200 plan) but eventually came back. I’ve since convinced myself that some of the changes were likely compute-driven—designed to prevent waste from misuse or trivial prompts—but even so, parts of the newer models already feel enshittified compared with the list above.
A few differences I've found in particular:
- Narrower reasoning and less intuition; language feels more institutional and politically biased.
- Weaker grasp of non-idiomatic English.
- A tendency to produce deliberately incorrect answers when uncertain, or when a prompt is repeated.
- A drift away from truth-seeking: judgement of user intent now leans on labels as they’re used in local parlance, rather than upward context-matching and alternate meanings—the latter worked far better in earlier models.
- A new fondness for flowery adjectives. Sonnet 3.7 never told me my code was “production-ready” or “beautiful.” Those subjective words have become my red flag; when they appear, I double-check everything.
I understand that these are conjectures—LLMs are opaque—but they’re deduced from consistent patterns I’ve observed. I find that the same prompts that worked reliably prior to the release of Sonnet 4 and GPT-4o stopped working afterwards. Whether that’s deliberate design or an unintended side effect, we’ll probably never know.
While the world does need it, the demand isn’t unavoidable yet. Most people aren’t motivated enough to act even for their own long-term good, let alone for an abstract ideal of decentralization. From a behavioral-economics view, the blocker is misaligned motivation: a decentralized photo app is a convenience for the few who care about autonomy, not a tangible benefit for the average user.
What might change this is a new class of tools—open-source or paid—built for power users who want to steer their own information environment. Think of them as “choose-your-own-reality” browsers that mix resource-fetching and synthetic-media recycling to create a more self-curated web. That seems a more plausible path than a mass migration to a decentralized Instagram clone.
We’ll also see the big platforms fragment as large sub-groups become dissatisfied and peel off. The result won’t look like a single decentralized network, more like many semi-centralized ones—small, durable ecosystems that eventually cross their own chasms. Investor optimism about infinite platform growth feels misplaced; we may have passed peak consolidation. The next decade should be interesting.
ChatGPT – journaling, talking, planning.
Codex – framework and middleware-layer coding.
Claude Code – logic and application-level coding.
Anthropic models via OpenRouter + Cline – when the task is error-prone, tedious, or needs high fidelity; lower error rate in my experience, though pricier.
Cursor Agents – multi-file integration, boilerplate, and forking tasks.
Each fills a different slot in the workflow, so “best” depends on what kind of coding you’re doing.
What I'm working on:
Unwrangle.com - An API for developers to query SERP, PDP and reviews data from online retailers and marketplaces.
Current focus:
Ant-ban strategies for higher / lower cost throughput. Trying to identify constraints to calculate feasibility, both technical and financial. This may be slightly controversial here since many are averse to bots and scraping. I’ve actually increased per-request costs because I suspect scraping will become more restricted and less tolerated over time — the supply-side signals point that way.
Ideas I'm thinking about:
Since I'm steering away from the higher concurrency/low cost scraping option — the new ideas I'm thinking about are: increasing data granularity, retailer coverage, adding an MCP server to help users query and analyse the E-commerce data they're extracting with the APIs as well.
Background:
I’ve been building this solo from India for about four years. It began as freelancing, then became an API product around a year ago. Today, I have ~90 customers, including a few reputed startups in California. For me the hardest parts are social, not technical or financial — staying connected to US working culture can feel inverted from here. I’ve applied to YC a few times and might again.
My service only extracts public data major retailers, not indie sites, and deducts more credits for lower-traffic domains to offset load differences.
It would be great if there were reliable ways to distinguish good bots from bad ones — many actually improve discoverability and sales. I see this with affiliate shopping sites that depend on e-commerce data, though that impact is hard to trace directly.
The bad actors are the ones cloning sites or using data for manipulation and propaganda.
I may have been too harsh. I love capitalism, technology, and software—they’ve built a meritocratic world and given me the tools to build my own life.
AI and technology feel like my best friend, but also my worst enemy when they edge toward learned helplessness. That tension exists with anything we depend on: the closer we get, the more power it holds.
The relationship between user and technology is becoming deeply intimate as systems gain reach and control. It’s important to stay optimistic but skeptical—and to keep protesting everything—because the work is moving faster than our ability to register its consequences.
Reading back, I realise I drifted into more of a monologue than a conversation. I get carried away when I’m trying to reason things out in public. Still, I stand by the core point about balance and transparency in how we shape the web.
As the legal history around scraping shows, it’s almost always the smaller company that gets sued out of existence. Taking on OpenAI or Microsoft, as you suggest, isn’t realistic — even governments often struggle to hold them accountable.
And for the record, large companies regularly ignore robots.txt themselves: LinkedIn, Google, OpenAI, and plenty of others.
The reality is that it’s the big players who behave like the aggressors, shaping the rules and breaking them when convenient. Smaller developers aren’t the problem, they’re just easier to punish.
With peering bandwidth being freely distributed to ISPs and consumers being fed media and subsidised services up until their necks makes the counter argument smell of narrative control rather than technical or financial constraints
But as I’m growing older I’m learning that the tech industry is mostly politically driven and relies on truth obfuscation as explained by Peter Thiel rather than real empowerment
It’s facilitating accumulation of control and power at an unparalleled pace. If anything it’s proving to be more unjust than the feudal systems it promises to replace.
It’s fair to be angry at abuse and "aggressive bots", but it's important to remember most large platforms—including the ones being scraped—built their own products on scraping too.
I run an e-commerce-specific scraping API that helps developers access SERP, PDP, and reviews data. I've noticed the web already has unsaid balances: certain traffic patterns and techniques are tolerated, others clearly aren’t. Most sites handle reasonable, well-behaved crawlers just fine.
Platforms claim ownership of UGC and public data through dark patterns and narrative control. The current guidelines are a result of supplier convenience, and there are several cases where absolutely fundamental web services run by the largest companies in the world themselves breach those guidelines (including those funded by the fund running this site). We need standards that treat public data as a shared resource with predictable, ethical access for everyone, not just for those with scale or lobbying power.
The Internet isn’t possible without scraping. For all the sentiment against scraping public data, doing so remains legal and essential to a lot of the services we use everyday. I think setting guidelines and shaping the web for reduced friction aimed at fair usage rather than turning it political would be the right thing to do.
Me too, and as the number and maturity of my projects have grown, improving and maintaining them all together has become harder by a factor I haven’t encountered before